1 /* SPDX-License-Identifier: GPL-2.0-or-later */ 2 /* 3 * Definitions for the 'struct sk_buff' memory handlers. 4 * 5 * Authors: 6 * Alan Cox, <[email protected]> 7 * Florian La Roche, <[email protected]> 8 */ 9 10 #ifndef _LINUX_SKBUFF_H 11 #define _LINUX_SKBUFF_H 12 13 #include <linux/kernel.h> 14 #include <linux/compiler.h> 15 #include <linux/time.h> 16 #include <linux/bug.h> 17 #include <linux/bvec.h> 18 #include <linux/cache.h> 19 #include <linux/rbtree.h> 20 #include <linux/socket.h> 21 #include <linux/refcount.h> 22 23 #include <linux/atomic.h> 24 #include <asm/types.h> 25 #include <linux/spinlock.h> 26 #include <net/checksum.h> 27 #include <linux/rcupdate.h> 28 #include <linux/dma-mapping.h> 29 #include <linux/netdev_features.h> 30 #include <net/flow_dissector.h> 31 #include <linux/in6.h> 32 #include <linux/if_packet.h> 33 #include <linux/llist.h> 34 #include <net/flow.h> 35 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 36 #include <linux/netfilter/nf_conntrack_common.h> 37 #endif 38 #include <net/net_debug.h> 39 #include <net/dropreason-core.h> 40 #include <net/netmem.h> 41 42 /** 43 * DOC: skb checksums 44 * 45 * The interface for checksum offload between the stack and networking drivers 46 * is as follows... 47 * 48 * IP checksum related features 49 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 50 * 51 * Drivers advertise checksum offload capabilities in the features of a device. 52 * From the stack's point of view these are capabilities offered by the driver. 53 * A driver typically only advertises features that it is capable of offloading 54 * to its device. 55 * 56 * .. flat-table:: Checksum related device features 57 * :widths: 1 10 58 * 59 * * - %NETIF_F_HW_CSUM 60 * - The driver (or its device) is able to compute one 61 * IP (one's complement) checksum for any combination 62 * of protocols or protocol layering. The checksum is 63 * computed and set in a packet per the CHECKSUM_PARTIAL 64 * interface (see below). 65 * 66 * * - %NETIF_F_IP_CSUM 67 * - Driver (device) is only able to checksum plain 68 * TCP or UDP packets over IPv4. These are specifically 69 * unencapsulated packets of the form IPv4|TCP or 70 * IPv4|UDP where the Protocol field in the IPv4 header 71 * is TCP or UDP. The IPv4 header may contain IP options. 72 * This feature cannot be set in features for a device 73 * with NETIF_F_HW_CSUM also set. This feature is being 74 * DEPRECATED (see below). 75 * 76 * * - %NETIF_F_IPV6_CSUM 77 * - Driver (device) is only able to checksum plain 78 * TCP or UDP packets over IPv6. These are specifically 79 * unencapsulated packets of the form IPv6|TCP or 80 * IPv6|UDP where the Next Header field in the IPv6 81 * header is either TCP or UDP. IPv6 extension headers 82 * are not supported with this feature. This feature 83 * cannot be set in features for a device with 84 * NETIF_F_HW_CSUM also set. This feature is being 85 * DEPRECATED (see below). 86 * 87 * * - %NETIF_F_RXCSUM 88 * - Driver (device) performs receive checksum offload. 89 * This flag is only used to disable the RX checksum 90 * feature for a device. The stack will accept receive 91 * checksum indication in packets received on a device 92 * regardless of whether NETIF_F_RXCSUM is set. 93 * 94 * Checksumming of received packets by device 95 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 96 * 97 * Indication of checksum verification is set in &sk_buff.ip_summed. 98 * Possible values are: 99 * 100 * - %CHECKSUM_NONE 101 * 102 * Device did not checksum this packet e.g. due to lack of capabilities. 103 * The packet contains full (though not verified) checksum in packet but 104 * not in skb->csum. Thus, skb->csum is undefined in this case. 105 * 106 * - %CHECKSUM_UNNECESSARY 107 * 108 * The hardware you're dealing with doesn't calculate the full checksum 109 * (as in %CHECKSUM_COMPLETE), but it does parse headers and verify checksums 110 * for specific protocols. For such packets it will set %CHECKSUM_UNNECESSARY 111 * if their checksums are okay. &sk_buff.csum is still undefined in this case 112 * though. A driver or device must never modify the checksum field in the 113 * packet even if checksum is verified. 114 * 115 * %CHECKSUM_UNNECESSARY is applicable to following protocols: 116 * 117 * - TCP: IPv6 and IPv4. 118 * - UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a 119 * zero UDP checksum for either IPv4 or IPv6, the networking stack 120 * may perform further validation in this case. 121 * - GRE: only if the checksum is present in the header. 122 * - SCTP: indicates the CRC in SCTP header has been validated. 123 * - FCOE: indicates the CRC in FC frame has been validated. 124 * 125 * &sk_buff.csum_level indicates the number of consecutive checksums found in 126 * the packet minus one that have been verified as %CHECKSUM_UNNECESSARY. 127 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet 128 * and a device is able to verify the checksums for UDP (possibly zero), 129 * GRE (checksum flag is set) and TCP, &sk_buff.csum_level would be set to 130 * two. If the device were only able to verify the UDP checksum and not 131 * GRE, either because it doesn't support GRE checksum or because GRE 132 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is 133 * not considered in this case). 134 * 135 * - %CHECKSUM_COMPLETE 136 * 137 * This is the most generic way. The device supplied checksum of the _whole_ 138 * packet as seen by netif_rx() and fills in &sk_buff.csum. This means the 139 * hardware doesn't need to parse L3/L4 headers to implement this. 140 * 141 * Notes: 142 * 143 * - Even if device supports only some protocols, but is able to produce 144 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY. 145 * - CHECKSUM_COMPLETE is not applicable to SCTP and FCoE protocols. 146 * 147 * - %CHECKSUM_PARTIAL 148 * 149 * A checksum is set up to be offloaded to a device as described in the 150 * output description for CHECKSUM_PARTIAL. This may occur on a packet 151 * received directly from another Linux OS, e.g., a virtualized Linux kernel 152 * on the same host, or it may be set in the input path in GRO or remote 153 * checksum offload. For the purposes of checksum verification, the checksum 154 * referred to by skb->csum_start + skb->csum_offset and any preceding 155 * checksums in the packet are considered verified. Any checksums in the 156 * packet that are after the checksum being offloaded are not considered to 157 * be verified. 158 * 159 * Checksumming on transmit for non-GSO 160 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 161 * 162 * The stack requests checksum offload in the &sk_buff.ip_summed for a packet. 163 * Values are: 164 * 165 * - %CHECKSUM_PARTIAL 166 * 167 * The driver is required to checksum the packet as seen by hard_start_xmit() 168 * from &sk_buff.csum_start up to the end, and to record/write the checksum at 169 * offset &sk_buff.csum_start + &sk_buff.csum_offset. 170 * A driver may verify that the 171 * csum_start and csum_offset values are valid values given the length and 172 * offset of the packet, but it should not attempt to validate that the 173 * checksum refers to a legitimate transport layer checksum -- it is the 174 * purview of the stack to validate that csum_start and csum_offset are set 175 * correctly. 176 * 177 * When the stack requests checksum offload for a packet, the driver MUST 178 * ensure that the checksum is set correctly. A driver can either offload the 179 * checksum calculation to the device, or call skb_checksum_help (in the case 180 * that the device does not support offload for a particular checksum). 181 * 182 * %NETIF_F_IP_CSUM and %NETIF_F_IPV6_CSUM are being deprecated in favor of 183 * %NETIF_F_HW_CSUM. New devices should use %NETIF_F_HW_CSUM to indicate 184 * checksum offload capability. 185 * skb_csum_hwoffload_help() can be called to resolve %CHECKSUM_PARTIAL based 186 * on network device checksumming capabilities: if a packet does not match 187 * them, skb_checksum_help() or skb_crc32c_help() (depending on the value of 188 * &sk_buff.csum_not_inet, see :ref:`crc`) 189 * is called to resolve the checksum. 190 * 191 * - %CHECKSUM_NONE 192 * 193 * The skb was already checksummed by the protocol, or a checksum is not 194 * required. 195 * 196 * - %CHECKSUM_UNNECESSARY 197 * 198 * This has the same meaning as CHECKSUM_NONE for checksum offload on 199 * output. 200 * 201 * - %CHECKSUM_COMPLETE 202 * 203 * Not used in checksum output. If a driver observes a packet with this value 204 * set in skbuff, it should treat the packet as if %CHECKSUM_NONE were set. 205 * 206 * .. _crc: 207 * 208 * Non-IP checksum (CRC) offloads 209 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 210 * 211 * .. flat-table:: 212 * :widths: 1 10 213 * 214 * * - %NETIF_F_SCTP_CRC 215 * - This feature indicates that a device is capable of 216 * offloading the SCTP CRC in a packet. To perform this offload the stack 217 * will set csum_start and csum_offset accordingly, set ip_summed to 218 * %CHECKSUM_PARTIAL and set csum_not_inet to 1, to provide an indication 219 * in the skbuff that the %CHECKSUM_PARTIAL refers to CRC32c. 220 * A driver that supports both IP checksum offload and SCTP CRC32c offload 221 * must verify which offload is configured for a packet by testing the 222 * value of &sk_buff.csum_not_inet; skb_crc32c_csum_help() is provided to 223 * resolve %CHECKSUM_PARTIAL on skbs where csum_not_inet is set to 1. 224 * 225 * * - %NETIF_F_FCOE_CRC 226 * - This feature indicates that a device is capable of offloading the FCOE 227 * CRC in a packet. To perform this offload the stack will set ip_summed 228 * to %CHECKSUM_PARTIAL and set csum_start and csum_offset 229 * accordingly. Note that there is no indication in the skbuff that the 230 * %CHECKSUM_PARTIAL refers to an FCOE checksum, so a driver that supports 231 * both IP checksum offload and FCOE CRC offload must verify which offload 232 * is configured for a packet, presumably by inspecting packet headers. 233 * 234 * Checksumming on output with GSO 235 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 236 * 237 * In the case of a GSO packet (skb_is_gso() is true), checksum offload 238 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the 239 * gso_type is %SKB_GSO_TCPV4 or %SKB_GSO_TCPV6, TCP checksum offload as 240 * part of the GSO operation is implied. If a checksum is being offloaded 241 * with GSO then ip_summed is %CHECKSUM_PARTIAL, and both csum_start and 242 * csum_offset are set to refer to the outermost checksum being offloaded 243 * (two offloaded checksums are possible with UDP encapsulation). 244 */ 245 246 /* Don't change this without changing skb_csum_unnecessary! */ 247 #define CHECKSUM_NONE 0 248 #define CHECKSUM_UNNECESSARY 1 249 #define CHECKSUM_COMPLETE 2 250 #define CHECKSUM_PARTIAL 3 251 252 /* Maximum value in skb->csum_level */ 253 #define SKB_MAX_CSUM_LEVEL 3 254 255 #define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES) 256 #define SKB_WITH_OVERHEAD(X) \ 257 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 258 259 /* For X bytes available in skb->head, what is the minimal 260 * allocation needed, knowing struct skb_shared_info needs 261 * to be aligned. 262 */ 263 #define SKB_HEAD_ALIGN(X) (SKB_DATA_ALIGN(X) + \ 264 SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 265 266 #define SKB_MAX_ORDER(X, ORDER) \ 267 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X)) 268 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0)) 269 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2)) 270 271 /* return minimum truesize of one skb containing X bytes of data */ 272 #define SKB_TRUESIZE(X) ((X) + \ 273 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \ 274 SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 275 276 struct ahash_request; 277 struct net_device; 278 struct scatterlist; 279 struct pipe_inode_info; 280 struct iov_iter; 281 struct napi_struct; 282 struct bpf_prog; 283 union bpf_attr; 284 struct skb_ext; 285 struct ts_config; 286 287 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 288 struct nf_bridge_info { 289 enum { 290 BRNF_PROTO_UNCHANGED, 291 BRNF_PROTO_8021Q, 292 BRNF_PROTO_PPPOE 293 } orig_proto:8; 294 u8 pkt_otherhost:1; 295 u8 in_prerouting:1; 296 u8 bridged_dnat:1; 297 u8 sabotage_in_done:1; 298 __u16 frag_max_size; 299 int physinif; 300 301 /* always valid & non-NULL from FORWARD on, for physdev match */ 302 struct net_device *physoutdev; 303 union { 304 /* prerouting: detect dnat in orig/reply direction */ 305 __be32 ipv4_daddr; 306 struct in6_addr ipv6_daddr; 307 308 /* after prerouting + nat detected: store original source 309 * mac since neigh resolution overwrites it, only used while 310 * skb is out in neigh layer. 311 */ 312 char neigh_header[8]; 313 }; 314 }; 315 #endif 316 317 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 318 /* Chain in tc_skb_ext will be used to share the tc chain with 319 * ovs recirc_id. It will be set to the current chain by tc 320 * and read by ovs to recirc_id. 321 */ 322 struct tc_skb_ext { 323 union { 324 u64 act_miss_cookie; 325 __u32 chain; 326 }; 327 __u16 mru; 328 __u16 zone; 329 u8 post_ct:1; 330 u8 post_ct_snat:1; 331 u8 post_ct_dnat:1; 332 u8 act_miss:1; /* Set if act_miss_cookie is used */ 333 u8 l2_miss:1; /* Set by bridge upon FDB or MDB miss */ 334 }; 335 #endif 336 337 struct sk_buff_head { 338 /* These two members must be first to match sk_buff. */ 339 struct_group_tagged(sk_buff_list, list, 340 struct sk_buff *next; 341 struct sk_buff *prev; 342 ); 343 344 __u32 qlen; 345 spinlock_t lock; 346 }; 347 348 struct sk_buff; 349 350 #ifndef CONFIG_MAX_SKB_FRAGS 351 # define CONFIG_MAX_SKB_FRAGS 17 352 #endif 353 354 #define MAX_SKB_FRAGS CONFIG_MAX_SKB_FRAGS 355 356 /* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to 357 * segment using its current segmentation instead. 358 */ 359 #define GSO_BY_FRAGS 0xFFFF 360 361 typedef struct skb_frag { 362 netmem_ref netmem; 363 unsigned int len; 364 unsigned int offset; 365 } skb_frag_t; 366 367 /** 368 * skb_frag_size() - Returns the size of a skb fragment 369 * @frag: skb fragment 370 */ 371 static inline unsigned int skb_frag_size(const skb_frag_t *frag) 372 { 373 return frag->len; 374 } 375 376 /** 377 * skb_frag_size_set() - Sets the size of a skb fragment 378 * @frag: skb fragment 379 * @size: size of fragment 380 */ 381 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size) 382 { 383 frag->len = size; 384 } 385 386 /** 387 * skb_frag_size_add() - Increments the size of a skb fragment by @delta 388 * @frag: skb fragment 389 * @delta: value to add 390 */ 391 static inline void skb_frag_size_add(skb_frag_t *frag, int delta) 392 { 393 frag->len += delta; 394 } 395 396 /** 397 * skb_frag_size_sub() - Decrements the size of a skb fragment by @delta 398 * @frag: skb fragment 399 * @delta: value to subtract 400 */ 401 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta) 402 { 403 frag->len -= delta; 404 } 405 406 /** 407 * skb_frag_must_loop - Test if %p is a high memory page 408 * @p: fragment's page 409 */ 410 static inline bool skb_frag_must_loop(struct page *p) 411 { 412 #if defined(CONFIG_HIGHMEM) 413 if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP) || PageHighMem(p)) 414 return true; 415 #endif 416 return false; 417 } 418 419 /** 420 * skb_frag_foreach_page - loop over pages in a fragment 421 * 422 * @f: skb frag to operate on 423 * @f_off: offset from start of f->netmem 424 * @f_len: length from f_off to loop over 425 * @p: (temp var) current page 426 * @p_off: (temp var) offset from start of current page, 427 * non-zero only on first page. 428 * @p_len: (temp var) length in current page, 429 * < PAGE_SIZE only on first and last page. 430 * @copied: (temp var) length so far, excluding current p_len. 431 * 432 * A fragment can hold a compound page, in which case per-page 433 * operations, notably kmap_atomic, must be called for each 434 * regular page. 435 */ 436 #define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \ 437 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \ 438 p_off = (f_off) & (PAGE_SIZE - 1), \ 439 p_len = skb_frag_must_loop(p) ? \ 440 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \ 441 copied = 0; \ 442 copied < f_len; \ 443 copied += p_len, p++, p_off = 0, \ 444 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \ 445 446 /** 447 * struct skb_shared_hwtstamps - hardware time stamps 448 * @hwtstamp: hardware time stamp transformed into duration 449 * since arbitrary point in time 450 * @netdev_data: address/cookie of network device driver used as 451 * reference to actual hardware time stamp 452 * 453 * Software time stamps generated by ktime_get_real() are stored in 454 * skb->tstamp. 455 * 456 * hwtstamps can only be compared against other hwtstamps from 457 * the same device. 458 * 459 * This structure is attached to packets as part of the 460 * &skb_shared_info. Use skb_hwtstamps() to get a pointer. 461 */ 462 struct skb_shared_hwtstamps { 463 union { 464 ktime_t hwtstamp; 465 void *netdev_data; 466 }; 467 }; 468 469 /* Definitions for tx_flags in struct skb_shared_info */ 470 enum { 471 /* generate hardware time stamp */ 472 SKBTX_HW_TSTAMP = 1 << 0, 473 474 /* generate software time stamp when queueing packet to NIC */ 475 SKBTX_SW_TSTAMP = 1 << 1, 476 477 /* device driver is going to provide hardware time stamp */ 478 SKBTX_IN_PROGRESS = 1 << 2, 479 480 /* generate hardware time stamp based on cycles if supported */ 481 SKBTX_HW_TSTAMP_USE_CYCLES = 1 << 3, 482 483 /* generate wifi status information (where possible) */ 484 SKBTX_WIFI_STATUS = 1 << 4, 485 486 /* determine hardware time stamp based on time or cycles */ 487 SKBTX_HW_TSTAMP_NETDEV = 1 << 5, 488 489 /* generate software time stamp when entering packet scheduling */ 490 SKBTX_SCHED_TSTAMP = 1 << 6, 491 }; 492 493 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \ 494 SKBTX_SCHED_TSTAMP) 495 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | \ 496 SKBTX_HW_TSTAMP_USE_CYCLES | \ 497 SKBTX_ANY_SW_TSTAMP) 498 499 /* Definitions for flags in struct skb_shared_info */ 500 enum { 501 /* use zcopy routines */ 502 SKBFL_ZEROCOPY_ENABLE = BIT(0), 503 504 /* This indicates at least one fragment might be overwritten 505 * (as in vmsplice(), sendfile() ...) 506 * If we need to compute a TX checksum, we'll need to copy 507 * all frags to avoid possible bad checksum 508 */ 509 SKBFL_SHARED_FRAG = BIT(1), 510 511 /* segment contains only zerocopy data and should not be 512 * charged to the kernel memory. 513 */ 514 SKBFL_PURE_ZEROCOPY = BIT(2), 515 516 SKBFL_DONT_ORPHAN = BIT(3), 517 518 /* page references are managed by the ubuf_info, so it's safe to 519 * use frags only up until ubuf_info is released 520 */ 521 SKBFL_MANAGED_FRAG_REFS = BIT(4), 522 }; 523 524 #define SKBFL_ZEROCOPY_FRAG (SKBFL_ZEROCOPY_ENABLE | SKBFL_SHARED_FRAG) 525 #define SKBFL_ALL_ZEROCOPY (SKBFL_ZEROCOPY_FRAG | SKBFL_PURE_ZEROCOPY | \ 526 SKBFL_DONT_ORPHAN | SKBFL_MANAGED_FRAG_REFS) 527 528 struct ubuf_info_ops { 529 void (*complete)(struct sk_buff *, struct ubuf_info *, 530 bool zerocopy_success); 531 /* has to be compatible with skb_zcopy_set() */ 532 int (*link_skb)(struct sk_buff *skb, struct ubuf_info *uarg); 533 }; 534 535 /* 536 * The callback notifies userspace to release buffers when skb DMA is done in 537 * lower device, the skb last reference should be 0 when calling this. 538 * The zerocopy_success argument is true if zero copy transmit occurred, 539 * false on data copy or out of memory error caused by data copy attempt. 540 * The ctx field is used to track device context. 541 * The desc field is used to track userspace buffer index. 542 */ 543 struct ubuf_info { 544 const struct ubuf_info_ops *ops; 545 refcount_t refcnt; 546 u8 flags; 547 }; 548 549 struct ubuf_info_msgzc { 550 struct ubuf_info ubuf; 551 552 union { 553 struct { 554 unsigned long desc; 555 void *ctx; 556 }; 557 struct { 558 u32 id; 559 u16 len; 560 u16 zerocopy:1; 561 u32 bytelen; 562 }; 563 }; 564 565 struct mmpin { 566 struct user_struct *user; 567 unsigned int num_pg; 568 } mmp; 569 }; 570 571 #define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg)) 572 #define uarg_to_msgzc(ubuf_ptr) container_of((ubuf_ptr), struct ubuf_info_msgzc, \ 573 ubuf) 574 575 int mm_account_pinned_pages(struct mmpin *mmp, size_t size); 576 void mm_unaccount_pinned_pages(struct mmpin *mmp); 577 578 /* Preserve some data across TX submission and completion. 579 * 580 * Note, this state is stored in the driver. Extending the layout 581 * might need some special care. 582 */ 583 struct xsk_tx_metadata_compl { 584 __u64 *tx_timestamp; 585 }; 586 587 /* This data is invariant across clones and lives at 588 * the end of the header data, ie. at skb->end. 589 */ 590 struct skb_shared_info { 591 __u8 flags; 592 __u8 meta_len; 593 __u8 nr_frags; 594 __u8 tx_flags; 595 unsigned short gso_size; 596 /* Warning: this field is not always filled in (UFO)! */ 597 unsigned short gso_segs; 598 struct sk_buff *frag_list; 599 union { 600 struct skb_shared_hwtstamps hwtstamps; 601 struct xsk_tx_metadata_compl xsk_meta; 602 }; 603 unsigned int gso_type; 604 u32 tskey; 605 606 /* 607 * Warning : all fields before dataref are cleared in __alloc_skb() 608 */ 609 atomic_t dataref; 610 unsigned int xdp_frags_size; 611 612 /* Intermediate layers must ensure that destructor_arg 613 * remains valid until skb destructor */ 614 void * destructor_arg; 615 616 /* must be last field, see pskb_expand_head() */ 617 skb_frag_t frags[MAX_SKB_FRAGS]; 618 }; 619 620 /** 621 * DOC: dataref and headerless skbs 622 * 623 * Transport layers send out clones of payload skbs they hold for 624 * retransmissions. To allow lower layers of the stack to prepend their headers 625 * we split &skb_shared_info.dataref into two halves. 626 * The lower 16 bits count the overall number of references. 627 * The higher 16 bits indicate how many of the references are payload-only. 628 * skb_header_cloned() checks if skb is allowed to add / write the headers. 629 * 630 * The creator of the skb (e.g. TCP) marks its skb as &sk_buff.nohdr 631 * (via __skb_header_release()). Any clone created from marked skb will get 632 * &sk_buff.hdr_len populated with the available headroom. 633 * If there's the only clone in existence it's able to modify the headroom 634 * at will. The sequence of calls inside the transport layer is:: 635 * 636 * <alloc skb> 637 * skb_reserve() 638 * __skb_header_release() 639 * skb_clone() 640 * // send the clone down the stack 641 * 642 * This is not a very generic construct and it depends on the transport layers 643 * doing the right thing. In practice there's usually only one payload-only skb. 644 * Having multiple payload-only skbs with different lengths of hdr_len is not 645 * possible. The payload-only skbs should never leave their owner. 646 */ 647 #define SKB_DATAREF_SHIFT 16 648 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1) 649 650 651 enum { 652 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */ 653 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */ 654 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */ 655 }; 656 657 enum { 658 SKB_GSO_TCPV4 = 1 << 0, 659 660 /* This indicates the skb is from an untrusted source. */ 661 SKB_GSO_DODGY = 1 << 1, 662 663 /* This indicates the tcp segment has CWR set. */ 664 SKB_GSO_TCP_ECN = 1 << 2, 665 666 SKB_GSO_TCP_FIXEDID = 1 << 3, 667 668 SKB_GSO_TCPV6 = 1 << 4, 669 670 SKB_GSO_FCOE = 1 << 5, 671 672 SKB_GSO_GRE = 1 << 6, 673 674 SKB_GSO_GRE_CSUM = 1 << 7, 675 676 SKB_GSO_IPXIP4 = 1 << 8, 677 678 SKB_GSO_IPXIP6 = 1 << 9, 679 680 SKB_GSO_UDP_TUNNEL = 1 << 10, 681 682 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11, 683 684 SKB_GSO_PARTIAL = 1 << 12, 685 686 SKB_GSO_TUNNEL_REMCSUM = 1 << 13, 687 688 SKB_GSO_SCTP = 1 << 14, 689 690 SKB_GSO_ESP = 1 << 15, 691 692 SKB_GSO_UDP = 1 << 16, 693 694 SKB_GSO_UDP_L4 = 1 << 17, 695 696 SKB_GSO_FRAGLIST = 1 << 18, 697 }; 698 699 #if BITS_PER_LONG > 32 700 #define NET_SKBUFF_DATA_USES_OFFSET 1 701 #endif 702 703 #ifdef NET_SKBUFF_DATA_USES_OFFSET 704 typedef unsigned int sk_buff_data_t; 705 #else 706 typedef unsigned char *sk_buff_data_t; 707 #endif 708 709 /** 710 * DOC: Basic sk_buff geometry 711 * 712 * struct sk_buff itself is a metadata structure and does not hold any packet 713 * data. All the data is held in associated buffers. 714 * 715 * &sk_buff.head points to the main "head" buffer. The head buffer is divided 716 * into two parts: 717 * 718 * - data buffer, containing headers and sometimes payload; 719 * this is the part of the skb operated on by the common helpers 720 * such as skb_put() or skb_pull(); 721 * - shared info (struct skb_shared_info) which holds an array of pointers 722 * to read-only data in the (page, offset, length) format. 723 * 724 * Optionally &skb_shared_info.frag_list may point to another skb. 725 * 726 * Basic diagram may look like this:: 727 * 728 * --------------- 729 * | sk_buff | 730 * --------------- 731 * ,--------------------------- + head 732 * / ,----------------- + data 733 * / / ,----------- + tail 734 * | | | , + end 735 * | | | | 736 * v v v v 737 * ----------------------------------------------- 738 * | headroom | data | tailroom | skb_shared_info | 739 * ----------------------------------------------- 740 * + [page frag] 741 * + [page frag] 742 * + [page frag] 743 * + [page frag] --------- 744 * + frag_list --> | sk_buff | 745 * --------- 746 * 747 */ 748 749 /** 750 * struct sk_buff - socket buffer 751 * @next: Next buffer in list 752 * @prev: Previous buffer in list 753 * @tstamp: Time we arrived/left 754 * @skb_mstamp_ns: (aka @tstamp) earliest departure time; start point 755 * for retransmit timer 756 * @rbnode: RB tree node, alternative to next/prev for netem/tcp 757 * @list: queue head 758 * @ll_node: anchor in an llist (eg socket defer_list) 759 * @sk: Socket we are owned by 760 * @dev: Device we arrived on/are leaving by 761 * @dev_scratch: (aka @dev) alternate use of @dev when @dev would be %NULL 762 * @cb: Control buffer. Free for use by every layer. Put private vars here 763 * @_skb_refdst: destination entry (with norefcount bit) 764 * @len: Length of actual data 765 * @data_len: Data length 766 * @mac_len: Length of link layer header 767 * @hdr_len: writable header length of cloned skb 768 * @csum: Checksum (must include start/offset pair) 769 * @csum_start: Offset from skb->head where checksumming should start 770 * @csum_offset: Offset from csum_start where checksum should be stored 771 * @priority: Packet queueing priority 772 * @ignore_df: allow local fragmentation 773 * @cloned: Head may be cloned (check refcnt to be sure) 774 * @ip_summed: Driver fed us an IP checksum 775 * @nohdr: Payload reference only, must not modify header 776 * @pkt_type: Packet class 777 * @fclone: skbuff clone status 778 * @ipvs_property: skbuff is owned by ipvs 779 * @inner_protocol_type: whether the inner protocol is 780 * ENCAP_TYPE_ETHER or ENCAP_TYPE_IPPROTO 781 * @remcsum_offload: remote checksum offload is enabled 782 * @offload_fwd_mark: Packet was L2-forwarded in hardware 783 * @offload_l3_fwd_mark: Packet was L3-forwarded in hardware 784 * @tc_skip_classify: do not classify packet. set by IFB device 785 * @tc_at_ingress: used within tc_classify to distinguish in/egress 786 * @redirected: packet was redirected by packet classifier 787 * @from_ingress: packet was redirected from the ingress path 788 * @nf_skip_egress: packet shall skip nf egress - see netfilter_netdev.h 789 * @peeked: this packet has been seen already, so stats have been 790 * done for it, don't do them again 791 * @nf_trace: netfilter packet trace flag 792 * @protocol: Packet protocol from driver 793 * @destructor: Destruct function 794 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue) 795 * @_sk_redir: socket redirection information for skmsg 796 * @_nfct: Associated connection, if any (with nfctinfo bits) 797 * @skb_iif: ifindex of device we arrived on 798 * @tc_index: Traffic control index 799 * @hash: the packet hash 800 * @queue_mapping: Queue mapping for multiqueue devices 801 * @head_frag: skb was allocated from page fragments, 802 * not allocated by kmalloc() or vmalloc(). 803 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves 804 * @pp_recycle: mark the packet for recycling instead of freeing (implies 805 * page_pool support on driver) 806 * @active_extensions: active extensions (skb_ext_id types) 807 * @ndisc_nodetype: router type (from link layer) 808 * @ooo_okay: allow the mapping of a socket to a queue to be changed 809 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport 810 * ports. 811 * @sw_hash: indicates hash was computed in software stack 812 * @wifi_acked_valid: wifi_acked was set 813 * @wifi_acked: whether frame was acked on wifi or not 814 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS 815 * @encapsulation: indicates the inner headers in the skbuff are valid 816 * @encap_hdr_csum: software checksum is needed 817 * @csum_valid: checksum is already valid 818 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL 819 * @csum_complete_sw: checksum was completed by software 820 * @csum_level: indicates the number of consecutive checksums found in 821 * the packet minus one that have been verified as 822 * CHECKSUM_UNNECESSARY (max 3) 823 * @dst_pending_confirm: need to confirm neighbour 824 * @decrypted: Decrypted SKB 825 * @slow_gro: state present at GRO time, slower prepare step required 826 * @mono_delivery_time: When set, skb->tstamp has the 827 * delivery_time in mono clock base (i.e. EDT). Otherwise, the 828 * skb->tstamp has the (rcv) timestamp at ingress and 829 * delivery_time at egress. 830 * @napi_id: id of the NAPI struct this skb came from 831 * @sender_cpu: (aka @napi_id) source CPU in XPS 832 * @alloc_cpu: CPU which did the skb allocation. 833 * @secmark: security marking 834 * @mark: Generic packet mark 835 * @reserved_tailroom: (aka @mark) number of bytes of free space available 836 * at the tail of an sk_buff 837 * @vlan_all: vlan fields (proto & tci) 838 * @vlan_proto: vlan encapsulation protocol 839 * @vlan_tci: vlan tag control information 840 * @inner_protocol: Protocol (encapsulation) 841 * @inner_ipproto: (aka @inner_protocol) stores ipproto when 842 * skb->inner_protocol_type == ENCAP_TYPE_IPPROTO; 843 * @inner_transport_header: Inner transport layer header (encapsulation) 844 * @inner_network_header: Network layer header (encapsulation) 845 * @inner_mac_header: Link layer header (encapsulation) 846 * @transport_header: Transport layer header 847 * @network_header: Network layer header 848 * @mac_header: Link layer header 849 * @kcov_handle: KCOV remote handle for remote coverage collection 850 * @tail: Tail pointer 851 * @end: End pointer 852 * @head: Head of buffer 853 * @data: Data head pointer 854 * @truesize: Buffer size 855 * @users: User count - see {datagram,tcp}.c 856 * @extensions: allocated extensions, valid if active_extensions is nonzero 857 */ 858 859 struct sk_buff { 860 union { 861 struct { 862 /* These two members must be first to match sk_buff_head. */ 863 struct sk_buff *next; 864 struct sk_buff *prev; 865 866 union { 867 struct net_device *dev; 868 /* Some protocols might use this space to store information, 869 * while device pointer would be NULL. 870 * UDP receive path is one user. 871 */ 872 unsigned long dev_scratch; 873 }; 874 }; 875 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */ 876 struct list_head list; 877 struct llist_node ll_node; 878 }; 879 880 struct sock *sk; 881 882 union { 883 ktime_t tstamp; 884 u64 skb_mstamp_ns; /* earliest departure time */ 885 }; 886 /* 887 * This is the control buffer. It is free to use for every 888 * layer. Please put your private variables there. If you 889 * want to keep them across layers you have to do a skb_clone() 890 * first. This is owned by whoever has the skb queued ATM. 891 */ 892 char cb[48] __aligned(8); 893 894 union { 895 struct { 896 unsigned long _skb_refdst; 897 void (*destructor)(struct sk_buff *skb); 898 }; 899 struct list_head tcp_tsorted_anchor; 900 #ifdef CONFIG_NET_SOCK_MSG 901 unsigned long _sk_redir; 902 #endif 903 }; 904 905 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 906 unsigned long _nfct; 907 #endif 908 unsigned int len, 909 data_len; 910 __u16 mac_len, 911 hdr_len; 912 913 /* Following fields are _not_ copied in __copy_skb_header() 914 * Note that queue_mapping is here mostly to fill a hole. 915 */ 916 __u16 queue_mapping; 917 918 /* if you move cloned around you also must adapt those constants */ 919 #ifdef __BIG_ENDIAN_BITFIELD 920 #define CLONED_MASK (1 << 7) 921 #else 922 #define CLONED_MASK 1 923 #endif 924 #define CLONED_OFFSET offsetof(struct sk_buff, __cloned_offset) 925 926 /* private: */ 927 __u8 __cloned_offset[0]; 928 /* public: */ 929 __u8 cloned:1, 930 nohdr:1, 931 fclone:2, 932 peeked:1, 933 head_frag:1, 934 pfmemalloc:1, 935 pp_recycle:1; /* page_pool recycle indicator */ 936 #ifdef CONFIG_SKB_EXTENSIONS 937 __u8 active_extensions; 938 #endif 939 940 /* Fields enclosed in headers group are copied 941 * using a single memcpy() in __copy_skb_header() 942 */ 943 struct_group(headers, 944 945 /* private: */ 946 __u8 __pkt_type_offset[0]; 947 /* public: */ 948 __u8 pkt_type:3; /* see PKT_TYPE_MAX */ 949 __u8 ignore_df:1; 950 __u8 dst_pending_confirm:1; 951 __u8 ip_summed:2; 952 __u8 ooo_okay:1; 953 954 /* private: */ 955 __u8 __mono_tc_offset[0]; 956 /* public: */ 957 __u8 mono_delivery_time:1; /* See SKB_MONO_DELIVERY_TIME_MASK */ 958 #ifdef CONFIG_NET_XGRESS 959 __u8 tc_at_ingress:1; /* See TC_AT_INGRESS_MASK */ 960 __u8 tc_skip_classify:1; 961 #endif 962 __u8 remcsum_offload:1; 963 __u8 csum_complete_sw:1; 964 __u8 csum_level:2; 965 __u8 inner_protocol_type:1; 966 967 __u8 l4_hash:1; 968 __u8 sw_hash:1; 969 #ifdef CONFIG_WIRELESS 970 __u8 wifi_acked_valid:1; 971 __u8 wifi_acked:1; 972 #endif 973 __u8 no_fcs:1; 974 /* Indicates the inner headers are valid in the skbuff. */ 975 __u8 encapsulation:1; 976 __u8 encap_hdr_csum:1; 977 __u8 csum_valid:1; 978 #ifdef CONFIG_IPV6_NDISC_NODETYPE 979 __u8 ndisc_nodetype:2; 980 #endif 981 982 #if IS_ENABLED(CONFIG_IP_VS) 983 __u8 ipvs_property:1; 984 #endif 985 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES) 986 __u8 nf_trace:1; 987 #endif 988 #ifdef CONFIG_NET_SWITCHDEV 989 __u8 offload_fwd_mark:1; 990 __u8 offload_l3_fwd_mark:1; 991 #endif 992 __u8 redirected:1; 993 #ifdef CONFIG_NET_REDIRECT 994 __u8 from_ingress:1; 995 #endif 996 #ifdef CONFIG_NETFILTER_SKIP_EGRESS 997 __u8 nf_skip_egress:1; 998 #endif 999 #ifdef CONFIG_SKB_DECRYPTED 1000 __u8 decrypted:1; 1001 #endif 1002 __u8 slow_gro:1; 1003 #if IS_ENABLED(CONFIG_IP_SCTP) 1004 __u8 csum_not_inet:1; 1005 #endif 1006 1007 #if defined(CONFIG_NET_SCHED) || defined(CONFIG_NET_XGRESS) 1008 __u16 tc_index; /* traffic control index */ 1009 #endif 1010 1011 u16 alloc_cpu; 1012 1013 union { 1014 __wsum csum; 1015 struct { 1016 __u16 csum_start; 1017 __u16 csum_offset; 1018 }; 1019 }; 1020 __u32 priority; 1021 int skb_iif; 1022 __u32 hash; 1023 union { 1024 u32 vlan_all; 1025 struct { 1026 __be16 vlan_proto; 1027 __u16 vlan_tci; 1028 }; 1029 }; 1030 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS) 1031 union { 1032 unsigned int napi_id; 1033 unsigned int sender_cpu; 1034 }; 1035 #endif 1036 #ifdef CONFIG_NETWORK_SECMARK 1037 __u32 secmark; 1038 #endif 1039 1040 union { 1041 __u32 mark; 1042 __u32 reserved_tailroom; 1043 }; 1044 1045 union { 1046 __be16 inner_protocol; 1047 __u8 inner_ipproto; 1048 }; 1049 1050 __u16 inner_transport_header; 1051 __u16 inner_network_header; 1052 __u16 inner_mac_header; 1053 1054 __be16 protocol; 1055 __u16 transport_header; 1056 __u16 network_header; 1057 __u16 mac_header; 1058 1059 #ifdef CONFIG_KCOV 1060 u64 kcov_handle; 1061 #endif 1062 1063 ); /* end headers group */ 1064 1065 /* These elements must be at the end, see alloc_skb() for details. */ 1066 sk_buff_data_t tail; 1067 sk_buff_data_t end; 1068 unsigned char *head, 1069 *data; 1070 unsigned int truesize; 1071 refcount_t users; 1072 1073 #ifdef CONFIG_SKB_EXTENSIONS 1074 /* only usable after checking ->active_extensions != 0 */ 1075 struct skb_ext *extensions; 1076 #endif 1077 }; 1078 1079 /* if you move pkt_type around you also must adapt those constants */ 1080 #ifdef __BIG_ENDIAN_BITFIELD 1081 #define PKT_TYPE_MAX (7 << 5) 1082 #else 1083 #define PKT_TYPE_MAX 7 1084 #endif 1085 #define PKT_TYPE_OFFSET offsetof(struct sk_buff, __pkt_type_offset) 1086 1087 /* if you move tc_at_ingress or mono_delivery_time 1088 * around, you also must adapt these constants. 1089 */ 1090 #ifdef __BIG_ENDIAN_BITFIELD 1091 #define SKB_MONO_DELIVERY_TIME_MASK (1 << 7) 1092 #define TC_AT_INGRESS_MASK (1 << 6) 1093 #else 1094 #define SKB_MONO_DELIVERY_TIME_MASK (1 << 0) 1095 #define TC_AT_INGRESS_MASK (1 << 1) 1096 #endif 1097 #define SKB_BF_MONO_TC_OFFSET offsetof(struct sk_buff, __mono_tc_offset) 1098 1099 #ifdef __KERNEL__ 1100 /* 1101 * Handling routines are only of interest to the kernel 1102 */ 1103 1104 #define SKB_ALLOC_FCLONE 0x01 1105 #define SKB_ALLOC_RX 0x02 1106 #define SKB_ALLOC_NAPI 0x04 1107 1108 /** 1109 * skb_pfmemalloc - Test if the skb was allocated from PFMEMALLOC reserves 1110 * @skb: buffer 1111 */ 1112 static inline bool skb_pfmemalloc(const struct sk_buff *skb) 1113 { 1114 return unlikely(skb->pfmemalloc); 1115 } 1116 1117 /* 1118 * skb might have a dst pointer attached, refcounted or not. 1119 * _skb_refdst low order bit is set if refcount was _not_ taken 1120 */ 1121 #define SKB_DST_NOREF 1UL 1122 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF) 1123 1124 /** 1125 * skb_dst - returns skb dst_entry 1126 * @skb: buffer 1127 * 1128 * Returns skb dst_entry, regardless of reference taken or not. 1129 */ 1130 static inline struct dst_entry *skb_dst(const struct sk_buff *skb) 1131 { 1132 /* If refdst was not refcounted, check we still are in a 1133 * rcu_read_lock section 1134 */ 1135 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) && 1136 !rcu_read_lock_held() && 1137 !rcu_read_lock_bh_held()); 1138 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK); 1139 } 1140 1141 /** 1142 * skb_dst_set - sets skb dst 1143 * @skb: buffer 1144 * @dst: dst entry 1145 * 1146 * Sets skb dst, assuming a reference was taken on dst and should 1147 * be released by skb_dst_drop() 1148 */ 1149 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst) 1150 { 1151 skb->slow_gro |= !!dst; 1152 skb->_skb_refdst = (unsigned long)dst; 1153 } 1154 1155 /** 1156 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference 1157 * @skb: buffer 1158 * @dst: dst entry 1159 * 1160 * Sets skb dst, assuming a reference was not taken on dst. 1161 * If dst entry is cached, we do not take reference and dst_release 1162 * will be avoided by refdst_drop. If dst entry is not cached, we take 1163 * reference, so that last dst_release can destroy the dst immediately. 1164 */ 1165 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst) 1166 { 1167 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 1168 skb->slow_gro |= !!dst; 1169 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF; 1170 } 1171 1172 /** 1173 * skb_dst_is_noref - Test if skb dst isn't refcounted 1174 * @skb: buffer 1175 */ 1176 static inline bool skb_dst_is_noref(const struct sk_buff *skb) 1177 { 1178 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb); 1179 } 1180 1181 /* For mangling skb->pkt_type from user space side from applications 1182 * such as nft, tc, etc, we only allow a conservative subset of 1183 * possible pkt_types to be set. 1184 */ 1185 static inline bool skb_pkt_type_ok(u32 ptype) 1186 { 1187 return ptype <= PACKET_OTHERHOST; 1188 } 1189 1190 /** 1191 * skb_napi_id - Returns the skb's NAPI id 1192 * @skb: buffer 1193 */ 1194 static inline unsigned int skb_napi_id(const struct sk_buff *skb) 1195 { 1196 #ifdef CONFIG_NET_RX_BUSY_POLL 1197 return skb->napi_id; 1198 #else 1199 return 0; 1200 #endif 1201 } 1202 1203 static inline bool skb_wifi_acked_valid(const struct sk_buff *skb) 1204 { 1205 #ifdef CONFIG_WIRELESS 1206 return skb->wifi_acked_valid; 1207 #else 1208 return 0; 1209 #endif 1210 } 1211 1212 /** 1213 * skb_unref - decrement the skb's reference count 1214 * @skb: buffer 1215 * 1216 * Returns true if we can free the skb. 1217 */ 1218 static inline bool skb_unref(struct sk_buff *skb) 1219 { 1220 if (unlikely(!skb)) 1221 return false; 1222 if (likely(refcount_read(&skb->users) == 1)) 1223 smp_rmb(); 1224 else if (likely(!refcount_dec_and_test(&skb->users))) 1225 return false; 1226 1227 return true; 1228 } 1229 1230 static inline bool skb_data_unref(const struct sk_buff *skb, 1231 struct skb_shared_info *shinfo) 1232 { 1233 int bias; 1234 1235 if (!skb->cloned) 1236 return true; 1237 1238 bias = skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1; 1239 1240 if (atomic_read(&shinfo->dataref) == bias) 1241 smp_rmb(); 1242 else if (atomic_sub_return(bias, &shinfo->dataref)) 1243 return false; 1244 1245 return true; 1246 } 1247 1248 void __fix_address 1249 kfree_skb_reason(struct sk_buff *skb, enum skb_drop_reason reason); 1250 1251 /** 1252 * kfree_skb - free an sk_buff with 'NOT_SPECIFIED' reason 1253 * @skb: buffer to free 1254 */ 1255 static inline void kfree_skb(struct sk_buff *skb) 1256 { 1257 kfree_skb_reason(skb, SKB_DROP_REASON_NOT_SPECIFIED); 1258 } 1259 1260 void skb_release_head_state(struct sk_buff *skb); 1261 void kfree_skb_list_reason(struct sk_buff *segs, 1262 enum skb_drop_reason reason); 1263 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt); 1264 void skb_tx_error(struct sk_buff *skb); 1265 1266 static inline void kfree_skb_list(struct sk_buff *segs) 1267 { 1268 kfree_skb_list_reason(segs, SKB_DROP_REASON_NOT_SPECIFIED); 1269 } 1270 1271 #ifdef CONFIG_TRACEPOINTS 1272 void consume_skb(struct sk_buff *skb); 1273 #else 1274 static inline void consume_skb(struct sk_buff *skb) 1275 { 1276 return kfree_skb(skb); 1277 } 1278 #endif 1279 1280 void __consume_stateless_skb(struct sk_buff *skb); 1281 void __kfree_skb(struct sk_buff *skb); 1282 1283 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen); 1284 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 1285 bool *fragstolen, int *delta_truesize); 1286 1287 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags, 1288 int node); 1289 struct sk_buff *__build_skb(void *data, unsigned int frag_size); 1290 struct sk_buff *build_skb(void *data, unsigned int frag_size); 1291 struct sk_buff *build_skb_around(struct sk_buff *skb, 1292 void *data, unsigned int frag_size); 1293 void skb_attempt_defer_free(struct sk_buff *skb); 1294 1295 struct sk_buff *napi_build_skb(void *data, unsigned int frag_size); 1296 struct sk_buff *slab_build_skb(void *data); 1297 1298 /** 1299 * alloc_skb - allocate a network buffer 1300 * @size: size to allocate 1301 * @priority: allocation mask 1302 * 1303 * This function is a convenient wrapper around __alloc_skb(). 1304 */ 1305 static inline struct sk_buff *alloc_skb(unsigned int size, 1306 gfp_t priority) 1307 { 1308 return __alloc_skb(size, priority, 0, NUMA_NO_NODE); 1309 } 1310 1311 struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 1312 unsigned long data_len, 1313 int max_page_order, 1314 int *errcode, 1315 gfp_t gfp_mask); 1316 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first); 1317 1318 /* Layout of fast clones : [skb1][skb2][fclone_ref] */ 1319 struct sk_buff_fclones { 1320 struct sk_buff skb1; 1321 1322 struct sk_buff skb2; 1323 1324 refcount_t fclone_ref; 1325 }; 1326 1327 /** 1328 * skb_fclone_busy - check if fclone is busy 1329 * @sk: socket 1330 * @skb: buffer 1331 * 1332 * Returns true if skb is a fast clone, and its clone is not freed. 1333 * Some drivers call skb_orphan() in their ndo_start_xmit(), 1334 * so we also check that didn't happen. 1335 */ 1336 static inline bool skb_fclone_busy(const struct sock *sk, 1337 const struct sk_buff *skb) 1338 { 1339 const struct sk_buff_fclones *fclones; 1340 1341 fclones = container_of(skb, struct sk_buff_fclones, skb1); 1342 1343 return skb->fclone == SKB_FCLONE_ORIG && 1344 refcount_read(&fclones->fclone_ref) > 1 && 1345 READ_ONCE(fclones->skb2.sk) == sk; 1346 } 1347 1348 /** 1349 * alloc_skb_fclone - allocate a network buffer from fclone cache 1350 * @size: size to allocate 1351 * @priority: allocation mask 1352 * 1353 * This function is a convenient wrapper around __alloc_skb(). 1354 */ 1355 static inline struct sk_buff *alloc_skb_fclone(unsigned int size, 1356 gfp_t priority) 1357 { 1358 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE); 1359 } 1360 1361 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src); 1362 void skb_headers_offset_update(struct sk_buff *skb, int off); 1363 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask); 1364 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority); 1365 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old); 1366 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority); 1367 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 1368 gfp_t gfp_mask, bool fclone); 1369 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, 1370 gfp_t gfp_mask) 1371 { 1372 return __pskb_copy_fclone(skb, headroom, gfp_mask, false); 1373 } 1374 1375 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask); 1376 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, 1377 unsigned int headroom); 1378 struct sk_buff *skb_expand_head(struct sk_buff *skb, unsigned int headroom); 1379 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom, 1380 int newtailroom, gfp_t priority); 1381 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 1382 int offset, int len); 1383 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, 1384 int offset, int len); 1385 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer); 1386 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error); 1387 1388 /** 1389 * skb_pad - zero pad the tail of an skb 1390 * @skb: buffer to pad 1391 * @pad: space to pad 1392 * 1393 * Ensure that a buffer is followed by a padding area that is zero 1394 * filled. Used by network drivers which may DMA or transfer data 1395 * beyond the buffer end onto the wire. 1396 * 1397 * May return error in out of memory cases. The skb is freed on error. 1398 */ 1399 static inline int skb_pad(struct sk_buff *skb, int pad) 1400 { 1401 return __skb_pad(skb, pad, true); 1402 } 1403 #define dev_kfree_skb(a) consume_skb(a) 1404 1405 int skb_append_pagefrags(struct sk_buff *skb, struct page *page, 1406 int offset, size_t size, size_t max_frags); 1407 1408 struct skb_seq_state { 1409 __u32 lower_offset; 1410 __u32 upper_offset; 1411 __u32 frag_idx; 1412 __u32 stepped_offset; 1413 struct sk_buff *root_skb; 1414 struct sk_buff *cur_skb; 1415 __u8 *frag_data; 1416 __u32 frag_off; 1417 }; 1418 1419 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 1420 unsigned int to, struct skb_seq_state *st); 1421 unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 1422 struct skb_seq_state *st); 1423 void skb_abort_seq_read(struct skb_seq_state *st); 1424 1425 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 1426 unsigned int to, struct ts_config *config); 1427 1428 /* 1429 * Packet hash types specify the type of hash in skb_set_hash. 1430 * 1431 * Hash types refer to the protocol layer addresses which are used to 1432 * construct a packet's hash. The hashes are used to differentiate or identify 1433 * flows of the protocol layer for the hash type. Hash types are either 1434 * layer-2 (L2), layer-3 (L3), or layer-4 (L4). 1435 * 1436 * Properties of hashes: 1437 * 1438 * 1) Two packets in different flows have different hash values 1439 * 2) Two packets in the same flow should have the same hash value 1440 * 1441 * A hash at a higher layer is considered to be more specific. A driver should 1442 * set the most specific hash possible. 1443 * 1444 * A driver cannot indicate a more specific hash than the layer at which a hash 1445 * was computed. For instance an L3 hash cannot be set as an L4 hash. 1446 * 1447 * A driver may indicate a hash level which is less specific than the 1448 * actual layer the hash was computed on. For instance, a hash computed 1449 * at L4 may be considered an L3 hash. This should only be done if the 1450 * driver can't unambiguously determine that the HW computed the hash at 1451 * the higher layer. Note that the "should" in the second property above 1452 * permits this. 1453 */ 1454 enum pkt_hash_types { 1455 PKT_HASH_TYPE_NONE, /* Undefined type */ 1456 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */ 1457 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */ 1458 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */ 1459 }; 1460 1461 static inline void skb_clear_hash(struct sk_buff *skb) 1462 { 1463 skb->hash = 0; 1464 skb->sw_hash = 0; 1465 skb->l4_hash = 0; 1466 } 1467 1468 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb) 1469 { 1470 if (!skb->l4_hash) 1471 skb_clear_hash(skb); 1472 } 1473 1474 static inline void 1475 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4) 1476 { 1477 skb->l4_hash = is_l4; 1478 skb->sw_hash = is_sw; 1479 skb->hash = hash; 1480 } 1481 1482 static inline void 1483 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type) 1484 { 1485 /* Used by drivers to set hash from HW */ 1486 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4); 1487 } 1488 1489 static inline void 1490 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4) 1491 { 1492 __skb_set_hash(skb, hash, true, is_l4); 1493 } 1494 1495 void __skb_get_hash(struct sk_buff *skb); 1496 u32 __skb_get_hash_symmetric(const struct sk_buff *skb); 1497 u32 skb_get_poff(const struct sk_buff *skb); 1498 u32 __skb_get_poff(const struct sk_buff *skb, const void *data, 1499 const struct flow_keys_basic *keys, int hlen); 1500 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto, 1501 const void *data, int hlen_proto); 1502 1503 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb, 1504 int thoff, u8 ip_proto) 1505 { 1506 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0); 1507 } 1508 1509 void skb_flow_dissector_init(struct flow_dissector *flow_dissector, 1510 const struct flow_dissector_key *key, 1511 unsigned int key_count); 1512 1513 struct bpf_flow_dissector; 1514 u32 bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx, 1515 __be16 proto, int nhoff, int hlen, unsigned int flags); 1516 1517 bool __skb_flow_dissect(const struct net *net, 1518 const struct sk_buff *skb, 1519 struct flow_dissector *flow_dissector, 1520 void *target_container, const void *data, 1521 __be16 proto, int nhoff, int hlen, unsigned int flags); 1522 1523 static inline bool skb_flow_dissect(const struct sk_buff *skb, 1524 struct flow_dissector *flow_dissector, 1525 void *target_container, unsigned int flags) 1526 { 1527 return __skb_flow_dissect(NULL, skb, flow_dissector, 1528 target_container, NULL, 0, 0, 0, flags); 1529 } 1530 1531 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb, 1532 struct flow_keys *flow, 1533 unsigned int flags) 1534 { 1535 memset(flow, 0, sizeof(*flow)); 1536 return __skb_flow_dissect(NULL, skb, &flow_keys_dissector, 1537 flow, NULL, 0, 0, 0, flags); 1538 } 1539 1540 static inline bool 1541 skb_flow_dissect_flow_keys_basic(const struct net *net, 1542 const struct sk_buff *skb, 1543 struct flow_keys_basic *flow, 1544 const void *data, __be16 proto, 1545 int nhoff, int hlen, unsigned int flags) 1546 { 1547 memset(flow, 0, sizeof(*flow)); 1548 return __skb_flow_dissect(net, skb, &flow_keys_basic_dissector, flow, 1549 data, proto, nhoff, hlen, flags); 1550 } 1551 1552 void skb_flow_dissect_meta(const struct sk_buff *skb, 1553 struct flow_dissector *flow_dissector, 1554 void *target_container); 1555 1556 /* Gets a skb connection tracking info, ctinfo map should be a 1557 * map of mapsize to translate enum ip_conntrack_info states 1558 * to user states. 1559 */ 1560 void 1561 skb_flow_dissect_ct(const struct sk_buff *skb, 1562 struct flow_dissector *flow_dissector, 1563 void *target_container, 1564 u16 *ctinfo_map, size_t mapsize, 1565 bool post_ct, u16 zone); 1566 void 1567 skb_flow_dissect_tunnel_info(const struct sk_buff *skb, 1568 struct flow_dissector *flow_dissector, 1569 void *target_container); 1570 1571 void skb_flow_dissect_hash(const struct sk_buff *skb, 1572 struct flow_dissector *flow_dissector, 1573 void *target_container); 1574 1575 static inline __u32 skb_get_hash(struct sk_buff *skb) 1576 { 1577 if (!skb->l4_hash && !skb->sw_hash) 1578 __skb_get_hash(skb); 1579 1580 return skb->hash; 1581 } 1582 1583 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6) 1584 { 1585 if (!skb->l4_hash && !skb->sw_hash) { 1586 struct flow_keys keys; 1587 __u32 hash = __get_hash_from_flowi6(fl6, &keys); 1588 1589 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys)); 1590 } 1591 1592 return skb->hash; 1593 } 1594 1595 __u32 skb_get_hash_perturb(const struct sk_buff *skb, 1596 const siphash_key_t *perturb); 1597 1598 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb) 1599 { 1600 return skb->hash; 1601 } 1602 1603 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from) 1604 { 1605 to->hash = from->hash; 1606 to->sw_hash = from->sw_hash; 1607 to->l4_hash = from->l4_hash; 1608 }; 1609 1610 static inline int skb_cmp_decrypted(const struct sk_buff *skb1, 1611 const struct sk_buff *skb2) 1612 { 1613 #ifdef CONFIG_SKB_DECRYPTED 1614 return skb2->decrypted - skb1->decrypted; 1615 #else 1616 return 0; 1617 #endif 1618 } 1619 1620 static inline bool skb_is_decrypted(const struct sk_buff *skb) 1621 { 1622 #ifdef CONFIG_SKB_DECRYPTED 1623 return skb->decrypted; 1624 #else 1625 return false; 1626 #endif 1627 } 1628 1629 static inline void skb_copy_decrypted(struct sk_buff *to, 1630 const struct sk_buff *from) 1631 { 1632 #ifdef CONFIG_SKB_DECRYPTED 1633 to->decrypted = from->decrypted; 1634 #endif 1635 } 1636 1637 #ifdef NET_SKBUFF_DATA_USES_OFFSET 1638 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1639 { 1640 return skb->head + skb->end; 1641 } 1642 1643 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1644 { 1645 return skb->end; 1646 } 1647 1648 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset) 1649 { 1650 skb->end = offset; 1651 } 1652 #else 1653 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1654 { 1655 return skb->end; 1656 } 1657 1658 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1659 { 1660 return skb->end - skb->head; 1661 } 1662 1663 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset) 1664 { 1665 skb->end = skb->head + offset; 1666 } 1667 #endif 1668 1669 extern const struct ubuf_info_ops msg_zerocopy_ubuf_ops; 1670 1671 struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size, 1672 struct ubuf_info *uarg); 1673 1674 void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref); 1675 1676 int __zerocopy_sg_from_iter(struct msghdr *msg, struct sock *sk, 1677 struct sk_buff *skb, struct iov_iter *from, 1678 size_t length); 1679 1680 static inline int skb_zerocopy_iter_dgram(struct sk_buff *skb, 1681 struct msghdr *msg, int len) 1682 { 1683 return __zerocopy_sg_from_iter(msg, skb->sk, skb, &msg->msg_iter, len); 1684 } 1685 1686 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb, 1687 struct msghdr *msg, int len, 1688 struct ubuf_info *uarg); 1689 1690 /* Internal */ 1691 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB))) 1692 1693 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb) 1694 { 1695 return &skb_shinfo(skb)->hwtstamps; 1696 } 1697 1698 static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb) 1699 { 1700 bool is_zcopy = skb && skb_shinfo(skb)->flags & SKBFL_ZEROCOPY_ENABLE; 1701 1702 return is_zcopy ? skb_uarg(skb) : NULL; 1703 } 1704 1705 static inline bool skb_zcopy_pure(const struct sk_buff *skb) 1706 { 1707 return skb_shinfo(skb)->flags & SKBFL_PURE_ZEROCOPY; 1708 } 1709 1710 static inline bool skb_zcopy_managed(const struct sk_buff *skb) 1711 { 1712 return skb_shinfo(skb)->flags & SKBFL_MANAGED_FRAG_REFS; 1713 } 1714 1715 static inline bool skb_pure_zcopy_same(const struct sk_buff *skb1, 1716 const struct sk_buff *skb2) 1717 { 1718 return skb_zcopy_pure(skb1) == skb_zcopy_pure(skb2); 1719 } 1720 1721 static inline void net_zcopy_get(struct ubuf_info *uarg) 1722 { 1723 refcount_inc(&uarg->refcnt); 1724 } 1725 1726 static inline void skb_zcopy_init(struct sk_buff *skb, struct ubuf_info *uarg) 1727 { 1728 skb_shinfo(skb)->destructor_arg = uarg; 1729 skb_shinfo(skb)->flags |= uarg->flags; 1730 } 1731 1732 static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg, 1733 bool *have_ref) 1734 { 1735 if (skb && uarg && !skb_zcopy(skb)) { 1736 if (unlikely(have_ref && *have_ref)) 1737 *have_ref = false; 1738 else 1739 net_zcopy_get(uarg); 1740 skb_zcopy_init(skb, uarg); 1741 } 1742 } 1743 1744 static inline void skb_zcopy_set_nouarg(struct sk_buff *skb, void *val) 1745 { 1746 skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL); 1747 skb_shinfo(skb)->flags |= SKBFL_ZEROCOPY_FRAG; 1748 } 1749 1750 static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb) 1751 { 1752 return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL; 1753 } 1754 1755 static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb) 1756 { 1757 return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL); 1758 } 1759 1760 static inline void net_zcopy_put(struct ubuf_info *uarg) 1761 { 1762 if (uarg) 1763 uarg->ops->complete(NULL, uarg, true); 1764 } 1765 1766 static inline void net_zcopy_put_abort(struct ubuf_info *uarg, bool have_uref) 1767 { 1768 if (uarg) { 1769 if (uarg->ops == &msg_zerocopy_ubuf_ops) 1770 msg_zerocopy_put_abort(uarg, have_uref); 1771 else if (have_uref) 1772 net_zcopy_put(uarg); 1773 } 1774 } 1775 1776 /* Release a reference on a zerocopy structure */ 1777 static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy_success) 1778 { 1779 struct ubuf_info *uarg = skb_zcopy(skb); 1780 1781 if (uarg) { 1782 if (!skb_zcopy_is_nouarg(skb)) 1783 uarg->ops->complete(skb, uarg, zerocopy_success); 1784 1785 skb_shinfo(skb)->flags &= ~SKBFL_ALL_ZEROCOPY; 1786 } 1787 } 1788 1789 void __skb_zcopy_downgrade_managed(struct sk_buff *skb); 1790 1791 static inline void skb_zcopy_downgrade_managed(struct sk_buff *skb) 1792 { 1793 if (unlikely(skb_zcopy_managed(skb))) 1794 __skb_zcopy_downgrade_managed(skb); 1795 } 1796 1797 static inline void skb_mark_not_on_list(struct sk_buff *skb) 1798 { 1799 skb->next = NULL; 1800 } 1801 1802 static inline void skb_poison_list(struct sk_buff *skb) 1803 { 1804 #ifdef CONFIG_DEBUG_NET 1805 skb->next = SKB_LIST_POISON_NEXT; 1806 #endif 1807 } 1808 1809 /* Iterate through singly-linked GSO fragments of an skb. */ 1810 #define skb_list_walk_safe(first, skb, next_skb) \ 1811 for ((skb) = (first), (next_skb) = (skb) ? (skb)->next : NULL; (skb); \ 1812 (skb) = (next_skb), (next_skb) = (skb) ? (skb)->next : NULL) 1813 1814 static inline void skb_list_del_init(struct sk_buff *skb) 1815 { 1816 __list_del_entry(&skb->list); 1817 skb_mark_not_on_list(skb); 1818 } 1819 1820 /** 1821 * skb_queue_empty - check if a queue is empty 1822 * @list: queue head 1823 * 1824 * Returns true if the queue is empty, false otherwise. 1825 */ 1826 static inline int skb_queue_empty(const struct sk_buff_head *list) 1827 { 1828 return list->next == (const struct sk_buff *) list; 1829 } 1830 1831 /** 1832 * skb_queue_empty_lockless - check if a queue is empty 1833 * @list: queue head 1834 * 1835 * Returns true if the queue is empty, false otherwise. 1836 * This variant can be used in lockless contexts. 1837 */ 1838 static inline bool skb_queue_empty_lockless(const struct sk_buff_head *list) 1839 { 1840 return READ_ONCE(list->next) == (const struct sk_buff *) list; 1841 } 1842 1843 1844 /** 1845 * skb_queue_is_last - check if skb is the last entry in the queue 1846 * @list: queue head 1847 * @skb: buffer 1848 * 1849 * Returns true if @skb is the last buffer on the list. 1850 */ 1851 static inline bool skb_queue_is_last(const struct sk_buff_head *list, 1852 const struct sk_buff *skb) 1853 { 1854 return skb->next == (const struct sk_buff *) list; 1855 } 1856 1857 /** 1858 * skb_queue_is_first - check if skb is the first entry in the queue 1859 * @list: queue head 1860 * @skb: buffer 1861 * 1862 * Returns true if @skb is the first buffer on the list. 1863 */ 1864 static inline bool skb_queue_is_first(const struct sk_buff_head *list, 1865 const struct sk_buff *skb) 1866 { 1867 return skb->prev == (const struct sk_buff *) list; 1868 } 1869 1870 /** 1871 * skb_queue_next - return the next packet in the queue 1872 * @list: queue head 1873 * @skb: current buffer 1874 * 1875 * Return the next packet in @list after @skb. It is only valid to 1876 * call this if skb_queue_is_last() evaluates to false. 1877 */ 1878 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list, 1879 const struct sk_buff *skb) 1880 { 1881 /* This BUG_ON may seem severe, but if we just return then we 1882 * are going to dereference garbage. 1883 */ 1884 BUG_ON(skb_queue_is_last(list, skb)); 1885 return skb->next; 1886 } 1887 1888 /** 1889 * skb_queue_prev - return the prev packet in the queue 1890 * @list: queue head 1891 * @skb: current buffer 1892 * 1893 * Return the prev packet in @list before @skb. It is only valid to 1894 * call this if skb_queue_is_first() evaluates to false. 1895 */ 1896 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list, 1897 const struct sk_buff *skb) 1898 { 1899 /* This BUG_ON may seem severe, but if we just return then we 1900 * are going to dereference garbage. 1901 */ 1902 BUG_ON(skb_queue_is_first(list, skb)); 1903 return skb->prev; 1904 } 1905 1906 /** 1907 * skb_get - reference buffer 1908 * @skb: buffer to reference 1909 * 1910 * Makes another reference to a socket buffer and returns a pointer 1911 * to the buffer. 1912 */ 1913 static inline struct sk_buff *skb_get(struct sk_buff *skb) 1914 { 1915 refcount_inc(&skb->users); 1916 return skb; 1917 } 1918 1919 /* 1920 * If users == 1, we are the only owner and can avoid redundant atomic changes. 1921 */ 1922 1923 /** 1924 * skb_cloned - is the buffer a clone 1925 * @skb: buffer to check 1926 * 1927 * Returns true if the buffer was generated with skb_clone() and is 1928 * one of multiple shared copies of the buffer. Cloned buffers are 1929 * shared data so must not be written to under normal circumstances. 1930 */ 1931 static inline int skb_cloned(const struct sk_buff *skb) 1932 { 1933 return skb->cloned && 1934 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1; 1935 } 1936 1937 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri) 1938 { 1939 might_sleep_if(gfpflags_allow_blocking(pri)); 1940 1941 if (skb_cloned(skb)) 1942 return pskb_expand_head(skb, 0, 0, pri); 1943 1944 return 0; 1945 } 1946 1947 /* This variant of skb_unclone() makes sure skb->truesize 1948 * and skb_end_offset() are not changed, whenever a new skb->head is needed. 1949 * 1950 * Indeed there is no guarantee that ksize(kmalloc(X)) == ksize(kmalloc(X)) 1951 * when various debugging features are in place. 1952 */ 1953 int __skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri); 1954 static inline int skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri) 1955 { 1956 might_sleep_if(gfpflags_allow_blocking(pri)); 1957 1958 if (skb_cloned(skb)) 1959 return __skb_unclone_keeptruesize(skb, pri); 1960 return 0; 1961 } 1962 1963 /** 1964 * skb_header_cloned - is the header a clone 1965 * @skb: buffer to check 1966 * 1967 * Returns true if modifying the header part of the buffer requires 1968 * the data to be copied. 1969 */ 1970 static inline int skb_header_cloned(const struct sk_buff *skb) 1971 { 1972 int dataref; 1973 1974 if (!skb->cloned) 1975 return 0; 1976 1977 dataref = atomic_read(&skb_shinfo(skb)->dataref); 1978 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT); 1979 return dataref != 1; 1980 } 1981 1982 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri) 1983 { 1984 might_sleep_if(gfpflags_allow_blocking(pri)); 1985 1986 if (skb_header_cloned(skb)) 1987 return pskb_expand_head(skb, 0, 0, pri); 1988 1989 return 0; 1990 } 1991 1992 /** 1993 * __skb_header_release() - allow clones to use the headroom 1994 * @skb: buffer to operate on 1995 * 1996 * See "DOC: dataref and headerless skbs". 1997 */ 1998 static inline void __skb_header_release(struct sk_buff *skb) 1999 { 2000 skb->nohdr = 1; 2001 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT)); 2002 } 2003 2004 2005 /** 2006 * skb_shared - is the buffer shared 2007 * @skb: buffer to check 2008 * 2009 * Returns true if more than one person has a reference to this 2010 * buffer. 2011 */ 2012 static inline int skb_shared(const struct sk_buff *skb) 2013 { 2014 return refcount_read(&skb->users) != 1; 2015 } 2016 2017 /** 2018 * skb_share_check - check if buffer is shared and if so clone it 2019 * @skb: buffer to check 2020 * @pri: priority for memory allocation 2021 * 2022 * If the buffer is shared the buffer is cloned and the old copy 2023 * drops a reference. A new clone with a single reference is returned. 2024 * If the buffer is not shared the original buffer is returned. When 2025 * being called from interrupt status or with spinlocks held pri must 2026 * be GFP_ATOMIC. 2027 * 2028 * NULL is returned on a memory allocation failure. 2029 */ 2030 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri) 2031 { 2032 might_sleep_if(gfpflags_allow_blocking(pri)); 2033 if (skb_shared(skb)) { 2034 struct sk_buff *nskb = skb_clone(skb, pri); 2035 2036 if (likely(nskb)) 2037 consume_skb(skb); 2038 else 2039 kfree_skb(skb); 2040 skb = nskb; 2041 } 2042 return skb; 2043 } 2044 2045 /* 2046 * Copy shared buffers into a new sk_buff. We effectively do COW on 2047 * packets to handle cases where we have a local reader and forward 2048 * and a couple of other messy ones. The normal one is tcpdumping 2049 * a packet that's being forwarded. 2050 */ 2051 2052 /** 2053 * skb_unshare - make a copy of a shared buffer 2054 * @skb: buffer to check 2055 * @pri: priority for memory allocation 2056 * 2057 * If the socket buffer is a clone then this function creates a new 2058 * copy of the data, drops a reference count on the old copy and returns 2059 * the new copy with the reference count at 1. If the buffer is not a clone 2060 * the original buffer is returned. When called with a spinlock held or 2061 * from interrupt state @pri must be %GFP_ATOMIC 2062 * 2063 * %NULL is returned on a memory allocation failure. 2064 */ 2065 static inline struct sk_buff *skb_unshare(struct sk_buff *skb, 2066 gfp_t pri) 2067 { 2068 might_sleep_if(gfpflags_allow_blocking(pri)); 2069 if (skb_cloned(skb)) { 2070 struct sk_buff *nskb = skb_copy(skb, pri); 2071 2072 /* Free our shared copy */ 2073 if (likely(nskb)) 2074 consume_skb(skb); 2075 else 2076 kfree_skb(skb); 2077 skb = nskb; 2078 } 2079 return skb; 2080 } 2081 2082 /** 2083 * skb_peek - peek at the head of an &sk_buff_head 2084 * @list_: list to peek at 2085 * 2086 * Peek an &sk_buff. Unlike most other operations you _MUST_ 2087 * be careful with this one. A peek leaves the buffer on the 2088 * list and someone else may run off with it. You must hold 2089 * the appropriate locks or have a private queue to do this. 2090 * 2091 * Returns %NULL for an empty list or a pointer to the head element. 2092 * The reference count is not incremented and the reference is therefore 2093 * volatile. Use with caution. 2094 */ 2095 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_) 2096 { 2097 struct sk_buff *skb = list_->next; 2098 2099 if (skb == (struct sk_buff *)list_) 2100 skb = NULL; 2101 return skb; 2102 } 2103 2104 /** 2105 * __skb_peek - peek at the head of a non-empty &sk_buff_head 2106 * @list_: list to peek at 2107 * 2108 * Like skb_peek(), but the caller knows that the list is not empty. 2109 */ 2110 static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_) 2111 { 2112 return list_->next; 2113 } 2114 2115 /** 2116 * skb_peek_next - peek skb following the given one from a queue 2117 * @skb: skb to start from 2118 * @list_: list to peek at 2119 * 2120 * Returns %NULL when the end of the list is met or a pointer to the 2121 * next element. The reference count is not incremented and the 2122 * reference is therefore volatile. Use with caution. 2123 */ 2124 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb, 2125 const struct sk_buff_head *list_) 2126 { 2127 struct sk_buff *next = skb->next; 2128 2129 if (next == (struct sk_buff *)list_) 2130 next = NULL; 2131 return next; 2132 } 2133 2134 /** 2135 * skb_peek_tail - peek at the tail of an &sk_buff_head 2136 * @list_: list to peek at 2137 * 2138 * Peek an &sk_buff. Unlike most other operations you _MUST_ 2139 * be careful with this one. A peek leaves the buffer on the 2140 * list and someone else may run off with it. You must hold 2141 * the appropriate locks or have a private queue to do this. 2142 * 2143 * Returns %NULL for an empty list or a pointer to the tail element. 2144 * The reference count is not incremented and the reference is therefore 2145 * volatile. Use with caution. 2146 */ 2147 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_) 2148 { 2149 struct sk_buff *skb = READ_ONCE(list_->prev); 2150 2151 if (skb == (struct sk_buff *)list_) 2152 skb = NULL; 2153 return skb; 2154 2155 } 2156 2157 /** 2158 * skb_queue_len - get queue length 2159 * @list_: list to measure 2160 * 2161 * Return the length of an &sk_buff queue. 2162 */ 2163 static inline __u32 skb_queue_len(const struct sk_buff_head *list_) 2164 { 2165 return list_->qlen; 2166 } 2167 2168 /** 2169 * skb_queue_len_lockless - get queue length 2170 * @list_: list to measure 2171 * 2172 * Return the length of an &sk_buff queue. 2173 * This variant can be used in lockless contexts. 2174 */ 2175 static inline __u32 skb_queue_len_lockless(const struct sk_buff_head *list_) 2176 { 2177 return READ_ONCE(list_->qlen); 2178 } 2179 2180 /** 2181 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head 2182 * @list: queue to initialize 2183 * 2184 * This initializes only the list and queue length aspects of 2185 * an sk_buff_head object. This allows to initialize the list 2186 * aspects of an sk_buff_head without reinitializing things like 2187 * the spinlock. It can also be used for on-stack sk_buff_head 2188 * objects where the spinlock is known to not be used. 2189 */ 2190 static inline void __skb_queue_head_init(struct sk_buff_head *list) 2191 { 2192 list->prev = list->next = (struct sk_buff *)list; 2193 list->qlen = 0; 2194 } 2195 2196 /* 2197 * This function creates a split out lock class for each invocation; 2198 * this is needed for now since a whole lot of users of the skb-queue 2199 * infrastructure in drivers have different locking usage (in hardirq) 2200 * than the networking core (in softirq only). In the long run either the 2201 * network layer or drivers should need annotation to consolidate the 2202 * main types of usage into 3 classes. 2203 */ 2204 static inline void skb_queue_head_init(struct sk_buff_head *list) 2205 { 2206 spin_lock_init(&list->lock); 2207 __skb_queue_head_init(list); 2208 } 2209 2210 static inline void skb_queue_head_init_class(struct sk_buff_head *list, 2211 struct lock_class_key *class) 2212 { 2213 skb_queue_head_init(list); 2214 lockdep_set_class(&list->lock, class); 2215 } 2216 2217 /* 2218 * Insert an sk_buff on a list. 2219 * 2220 * The "__skb_xxxx()" functions are the non-atomic ones that 2221 * can only be called with interrupts disabled. 2222 */ 2223 static inline void __skb_insert(struct sk_buff *newsk, 2224 struct sk_buff *prev, struct sk_buff *next, 2225 struct sk_buff_head *list) 2226 { 2227 /* See skb_queue_empty_lockless() and skb_peek_tail() 2228 * for the opposite READ_ONCE() 2229 */ 2230 WRITE_ONCE(newsk->next, next); 2231 WRITE_ONCE(newsk->prev, prev); 2232 WRITE_ONCE(((struct sk_buff_list *)next)->prev, newsk); 2233 WRITE_ONCE(((struct sk_buff_list *)prev)->next, newsk); 2234 WRITE_ONCE(list->qlen, list->qlen + 1); 2235 } 2236 2237 static inline void __skb_queue_splice(const struct sk_buff_head *list, 2238 struct sk_buff *prev, 2239 struct sk_buff *next) 2240 { 2241 struct sk_buff *first = list->next; 2242 struct sk_buff *last = list->prev; 2243 2244 WRITE_ONCE(first->prev, prev); 2245 WRITE_ONCE(prev->next, first); 2246 2247 WRITE_ONCE(last->next, next); 2248 WRITE_ONCE(next->prev, last); 2249 } 2250 2251 /** 2252 * skb_queue_splice - join two skb lists, this is designed for stacks 2253 * @list: the new list to add 2254 * @head: the place to add it in the first list 2255 */ 2256 static inline void skb_queue_splice(const struct sk_buff_head *list, 2257 struct sk_buff_head *head) 2258 { 2259 if (!skb_queue_empty(list)) { 2260 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 2261 head->qlen += list->qlen; 2262 } 2263 } 2264 2265 /** 2266 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list 2267 * @list: the new list to add 2268 * @head: the place to add it in the first list 2269 * 2270 * The list at @list is reinitialised 2271 */ 2272 static inline void skb_queue_splice_init(struct sk_buff_head *list, 2273 struct sk_buff_head *head) 2274 { 2275 if (!skb_queue_empty(list)) { 2276 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 2277 head->qlen += list->qlen; 2278 __skb_queue_head_init(list); 2279 } 2280 } 2281 2282 /** 2283 * skb_queue_splice_tail - join two skb lists, each list being a queue 2284 * @list: the new list to add 2285 * @head: the place to add it in the first list 2286 */ 2287 static inline void skb_queue_splice_tail(const struct sk_buff_head *list, 2288 struct sk_buff_head *head) 2289 { 2290 if (!skb_queue_empty(list)) { 2291 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 2292 head->qlen += list->qlen; 2293 } 2294 } 2295 2296 /** 2297 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list 2298 * @list: the new list to add 2299 * @head: the place to add it in the first list 2300 * 2301 * Each of the lists is a queue. 2302 * The list at @list is reinitialised 2303 */ 2304 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list, 2305 struct sk_buff_head *head) 2306 { 2307 if (!skb_queue_empty(list)) { 2308 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 2309 head->qlen += list->qlen; 2310 __skb_queue_head_init(list); 2311 } 2312 } 2313 2314 /** 2315 * __skb_queue_after - queue a buffer at the list head 2316 * @list: list to use 2317 * @prev: place after this buffer 2318 * @newsk: buffer to queue 2319 * 2320 * Queue a buffer int the middle of a list. This function takes no locks 2321 * and you must therefore hold required locks before calling it. 2322 * 2323 * A buffer cannot be placed on two lists at the same time. 2324 */ 2325 static inline void __skb_queue_after(struct sk_buff_head *list, 2326 struct sk_buff *prev, 2327 struct sk_buff *newsk) 2328 { 2329 __skb_insert(newsk, prev, ((struct sk_buff_list *)prev)->next, list); 2330 } 2331 2332 void skb_append(struct sk_buff *old, struct sk_buff *newsk, 2333 struct sk_buff_head *list); 2334 2335 static inline void __skb_queue_before(struct sk_buff_head *list, 2336 struct sk_buff *next, 2337 struct sk_buff *newsk) 2338 { 2339 __skb_insert(newsk, ((struct sk_buff_list *)next)->prev, next, list); 2340 } 2341 2342 /** 2343 * __skb_queue_head - queue a buffer at the list head 2344 * @list: list to use 2345 * @newsk: buffer to queue 2346 * 2347 * Queue a buffer at the start of a list. This function takes no locks 2348 * and you must therefore hold required locks before calling it. 2349 * 2350 * A buffer cannot be placed on two lists at the same time. 2351 */ 2352 static inline void __skb_queue_head(struct sk_buff_head *list, 2353 struct sk_buff *newsk) 2354 { 2355 __skb_queue_after(list, (struct sk_buff *)list, newsk); 2356 } 2357 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk); 2358 2359 /** 2360 * __skb_queue_tail - queue a buffer at the list tail 2361 * @list: list to use 2362 * @newsk: buffer to queue 2363 * 2364 * Queue a buffer at the end of a list. This function takes no locks 2365 * and you must therefore hold required locks before calling it. 2366 * 2367 * A buffer cannot be placed on two lists at the same time. 2368 */ 2369 static inline void __skb_queue_tail(struct sk_buff_head *list, 2370 struct sk_buff *newsk) 2371 { 2372 __skb_queue_before(list, (struct sk_buff *)list, newsk); 2373 } 2374 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk); 2375 2376 /* 2377 * remove sk_buff from list. _Must_ be called atomically, and with 2378 * the list known.. 2379 */ 2380 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list); 2381 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 2382 { 2383 struct sk_buff *next, *prev; 2384 2385 WRITE_ONCE(list->qlen, list->qlen - 1); 2386 next = skb->next; 2387 prev = skb->prev; 2388 skb->next = skb->prev = NULL; 2389 WRITE_ONCE(next->prev, prev); 2390 WRITE_ONCE(prev->next, next); 2391 } 2392 2393 /** 2394 * __skb_dequeue - remove from the head of the queue 2395 * @list: list to dequeue from 2396 * 2397 * Remove the head of the list. This function does not take any locks 2398 * so must be used with appropriate locks held only. The head item is 2399 * returned or %NULL if the list is empty. 2400 */ 2401 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list) 2402 { 2403 struct sk_buff *skb = skb_peek(list); 2404 if (skb) 2405 __skb_unlink(skb, list); 2406 return skb; 2407 } 2408 struct sk_buff *skb_dequeue(struct sk_buff_head *list); 2409 2410 /** 2411 * __skb_dequeue_tail - remove from the tail of the queue 2412 * @list: list to dequeue from 2413 * 2414 * Remove the tail of the list. This function does not take any locks 2415 * so must be used with appropriate locks held only. The tail item is 2416 * returned or %NULL if the list is empty. 2417 */ 2418 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list) 2419 { 2420 struct sk_buff *skb = skb_peek_tail(list); 2421 if (skb) 2422 __skb_unlink(skb, list); 2423 return skb; 2424 } 2425 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list); 2426 2427 2428 static inline bool skb_is_nonlinear(const struct sk_buff *skb) 2429 { 2430 return skb->data_len; 2431 } 2432 2433 static inline unsigned int skb_headlen(const struct sk_buff *skb) 2434 { 2435 return skb->len - skb->data_len; 2436 } 2437 2438 static inline unsigned int __skb_pagelen(const struct sk_buff *skb) 2439 { 2440 unsigned int i, len = 0; 2441 2442 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--) 2443 len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 2444 return len; 2445 } 2446 2447 static inline unsigned int skb_pagelen(const struct sk_buff *skb) 2448 { 2449 return skb_headlen(skb) + __skb_pagelen(skb); 2450 } 2451 2452 static inline void skb_frag_fill_netmem_desc(skb_frag_t *frag, 2453 netmem_ref netmem, int off, 2454 int size) 2455 { 2456 frag->netmem = netmem; 2457 frag->offset = off; 2458 skb_frag_size_set(frag, size); 2459 } 2460 2461 static inline void skb_frag_fill_page_desc(skb_frag_t *frag, 2462 struct page *page, 2463 int off, int size) 2464 { 2465 skb_frag_fill_netmem_desc(frag, page_to_netmem(page), off, size); 2466 } 2467 2468 static inline void __skb_fill_netmem_desc_noacc(struct skb_shared_info *shinfo, 2469 int i, netmem_ref netmem, 2470 int off, int size) 2471 { 2472 skb_frag_t *frag = &shinfo->frags[i]; 2473 2474 skb_frag_fill_netmem_desc(frag, netmem, off, size); 2475 } 2476 2477 static inline void __skb_fill_page_desc_noacc(struct skb_shared_info *shinfo, 2478 int i, struct page *page, 2479 int off, int size) 2480 { 2481 __skb_fill_netmem_desc_noacc(shinfo, i, page_to_netmem(page), off, 2482 size); 2483 } 2484 2485 /** 2486 * skb_len_add - adds a number to len fields of skb 2487 * @skb: buffer to add len to 2488 * @delta: number of bytes to add 2489 */ 2490 static inline void skb_len_add(struct sk_buff *skb, int delta) 2491 { 2492 skb->len += delta; 2493 skb->data_len += delta; 2494 skb->truesize += delta; 2495 } 2496 2497 /** 2498 * __skb_fill_netmem_desc - initialise a fragment in an skb 2499 * @skb: buffer containing fragment to be initialised 2500 * @i: fragment index to initialise 2501 * @netmem: the netmem to use for this fragment 2502 * @off: the offset to the data with @page 2503 * @size: the length of the data 2504 * 2505 * Initialises the @i'th fragment of @skb to point to &size bytes at 2506 * offset @off within @page. 2507 * 2508 * Does not take any additional reference on the fragment. 2509 */ 2510 static inline void __skb_fill_netmem_desc(struct sk_buff *skb, int i, 2511 netmem_ref netmem, int off, int size) 2512 { 2513 struct page *page = netmem_to_page(netmem); 2514 2515 __skb_fill_netmem_desc_noacc(skb_shinfo(skb), i, netmem, off, size); 2516 2517 /* Propagate page pfmemalloc to the skb if we can. The problem is 2518 * that not all callers have unique ownership of the page but rely 2519 * on page_is_pfmemalloc doing the right thing(tm). 2520 */ 2521 page = compound_head(page); 2522 if (page_is_pfmemalloc(page)) 2523 skb->pfmemalloc = true; 2524 } 2525 2526 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i, 2527 struct page *page, int off, int size) 2528 { 2529 __skb_fill_netmem_desc(skb, i, page_to_netmem(page), off, size); 2530 } 2531 2532 static inline void skb_fill_netmem_desc(struct sk_buff *skb, int i, 2533 netmem_ref netmem, int off, int size) 2534 { 2535 __skb_fill_netmem_desc(skb, i, netmem, off, size); 2536 skb_shinfo(skb)->nr_frags = i + 1; 2537 } 2538 2539 /** 2540 * skb_fill_page_desc - initialise a paged fragment in an skb 2541 * @skb: buffer containing fragment to be initialised 2542 * @i: paged fragment index to initialise 2543 * @page: the page to use for this fragment 2544 * @off: the offset to the data with @page 2545 * @size: the length of the data 2546 * 2547 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of 2548 * @skb to point to @size bytes at offset @off within @page. In 2549 * addition updates @skb such that @i is the last fragment. 2550 * 2551 * Does not take any additional reference on the fragment. 2552 */ 2553 static inline void skb_fill_page_desc(struct sk_buff *skb, int i, 2554 struct page *page, int off, int size) 2555 { 2556 skb_fill_netmem_desc(skb, i, page_to_netmem(page), off, size); 2557 } 2558 2559 /** 2560 * skb_fill_page_desc_noacc - initialise a paged fragment in an skb 2561 * @skb: buffer containing fragment to be initialised 2562 * @i: paged fragment index to initialise 2563 * @page: the page to use for this fragment 2564 * @off: the offset to the data with @page 2565 * @size: the length of the data 2566 * 2567 * Variant of skb_fill_page_desc() which does not deal with 2568 * pfmemalloc, if page is not owned by us. 2569 */ 2570 static inline void skb_fill_page_desc_noacc(struct sk_buff *skb, int i, 2571 struct page *page, int off, 2572 int size) 2573 { 2574 struct skb_shared_info *shinfo = skb_shinfo(skb); 2575 2576 __skb_fill_page_desc_noacc(shinfo, i, page, off, size); 2577 shinfo->nr_frags = i + 1; 2578 } 2579 2580 void skb_add_rx_frag_netmem(struct sk_buff *skb, int i, netmem_ref netmem, 2581 int off, int size, unsigned int truesize); 2582 2583 static inline void skb_add_rx_frag(struct sk_buff *skb, int i, 2584 struct page *page, int off, int size, 2585 unsigned int truesize) 2586 { 2587 skb_add_rx_frag_netmem(skb, i, page_to_netmem(page), off, size, 2588 truesize); 2589 } 2590 2591 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 2592 unsigned int truesize); 2593 2594 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb)) 2595 2596 #ifdef NET_SKBUFF_DATA_USES_OFFSET 2597 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 2598 { 2599 return skb->head + skb->tail; 2600 } 2601 2602 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 2603 { 2604 skb->tail = skb->data - skb->head; 2605 } 2606 2607 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 2608 { 2609 skb_reset_tail_pointer(skb); 2610 skb->tail += offset; 2611 } 2612 2613 #else /* NET_SKBUFF_DATA_USES_OFFSET */ 2614 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 2615 { 2616 return skb->tail; 2617 } 2618 2619 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 2620 { 2621 skb->tail = skb->data; 2622 } 2623 2624 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 2625 { 2626 skb->tail = skb->data + offset; 2627 } 2628 2629 #endif /* NET_SKBUFF_DATA_USES_OFFSET */ 2630 2631 static inline void skb_assert_len(struct sk_buff *skb) 2632 { 2633 #ifdef CONFIG_DEBUG_NET 2634 if (WARN_ONCE(!skb->len, "%s\n", __func__)) 2635 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 2636 #endif /* CONFIG_DEBUG_NET */ 2637 } 2638 2639 /* 2640 * Add data to an sk_buff 2641 */ 2642 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len); 2643 void *skb_put(struct sk_buff *skb, unsigned int len); 2644 static inline void *__skb_put(struct sk_buff *skb, unsigned int len) 2645 { 2646 void *tmp = skb_tail_pointer(skb); 2647 SKB_LINEAR_ASSERT(skb); 2648 skb->tail += len; 2649 skb->len += len; 2650 return tmp; 2651 } 2652 2653 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len) 2654 { 2655 void *tmp = __skb_put(skb, len); 2656 2657 memset(tmp, 0, len); 2658 return tmp; 2659 } 2660 2661 static inline void *__skb_put_data(struct sk_buff *skb, const void *data, 2662 unsigned int len) 2663 { 2664 void *tmp = __skb_put(skb, len); 2665 2666 memcpy(tmp, data, len); 2667 return tmp; 2668 } 2669 2670 static inline void __skb_put_u8(struct sk_buff *skb, u8 val) 2671 { 2672 *(u8 *)__skb_put(skb, 1) = val; 2673 } 2674 2675 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len) 2676 { 2677 void *tmp = skb_put(skb, len); 2678 2679 memset(tmp, 0, len); 2680 2681 return tmp; 2682 } 2683 2684 static inline void *skb_put_data(struct sk_buff *skb, const void *data, 2685 unsigned int len) 2686 { 2687 void *tmp = skb_put(skb, len); 2688 2689 memcpy(tmp, data, len); 2690 2691 return tmp; 2692 } 2693 2694 static inline void skb_put_u8(struct sk_buff *skb, u8 val) 2695 { 2696 *(u8 *)skb_put(skb, 1) = val; 2697 } 2698 2699 void *skb_push(struct sk_buff *skb, unsigned int len); 2700 static inline void *__skb_push(struct sk_buff *skb, unsigned int len) 2701 { 2702 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX); 2703 2704 skb->data -= len; 2705 skb->len += len; 2706 return skb->data; 2707 } 2708 2709 void *skb_pull(struct sk_buff *skb, unsigned int len); 2710 static inline void *__skb_pull(struct sk_buff *skb, unsigned int len) 2711 { 2712 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX); 2713 2714 skb->len -= len; 2715 if (unlikely(skb->len < skb->data_len)) { 2716 #if defined(CONFIG_DEBUG_NET) 2717 skb->len += len; 2718 pr_err("__skb_pull(len=%u)\n", len); 2719 skb_dump(KERN_ERR, skb, false); 2720 #endif 2721 BUG(); 2722 } 2723 return skb->data += len; 2724 } 2725 2726 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len) 2727 { 2728 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len); 2729 } 2730 2731 void *skb_pull_data(struct sk_buff *skb, size_t len); 2732 2733 void *__pskb_pull_tail(struct sk_buff *skb, int delta); 2734 2735 static inline enum skb_drop_reason 2736 pskb_may_pull_reason(struct sk_buff *skb, unsigned int len) 2737 { 2738 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX); 2739 2740 if (likely(len <= skb_headlen(skb))) 2741 return SKB_NOT_DROPPED_YET; 2742 2743 if (unlikely(len > skb->len)) 2744 return SKB_DROP_REASON_PKT_TOO_SMALL; 2745 2746 if (unlikely(!__pskb_pull_tail(skb, len - skb_headlen(skb)))) 2747 return SKB_DROP_REASON_NOMEM; 2748 2749 return SKB_NOT_DROPPED_YET; 2750 } 2751 2752 static inline bool pskb_may_pull(struct sk_buff *skb, unsigned int len) 2753 { 2754 return pskb_may_pull_reason(skb, len) == SKB_NOT_DROPPED_YET; 2755 } 2756 2757 static inline void *pskb_pull(struct sk_buff *skb, unsigned int len) 2758 { 2759 if (!pskb_may_pull(skb, len)) 2760 return NULL; 2761 2762 skb->len -= len; 2763 return skb->data += len; 2764 } 2765 2766 void skb_condense(struct sk_buff *skb); 2767 2768 /** 2769 * skb_headroom - bytes at buffer head 2770 * @skb: buffer to check 2771 * 2772 * Return the number of bytes of free space at the head of an &sk_buff. 2773 */ 2774 static inline unsigned int skb_headroom(const struct sk_buff *skb) 2775 { 2776 return skb->data - skb->head; 2777 } 2778 2779 /** 2780 * skb_tailroom - bytes at buffer end 2781 * @skb: buffer to check 2782 * 2783 * Return the number of bytes of free space at the tail of an sk_buff 2784 */ 2785 static inline int skb_tailroom(const struct sk_buff *skb) 2786 { 2787 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail; 2788 } 2789 2790 /** 2791 * skb_availroom - bytes at buffer end 2792 * @skb: buffer to check 2793 * 2794 * Return the number of bytes of free space at the tail of an sk_buff 2795 * allocated by sk_stream_alloc() 2796 */ 2797 static inline int skb_availroom(const struct sk_buff *skb) 2798 { 2799 if (skb_is_nonlinear(skb)) 2800 return 0; 2801 2802 return skb->end - skb->tail - skb->reserved_tailroom; 2803 } 2804 2805 /** 2806 * skb_reserve - adjust headroom 2807 * @skb: buffer to alter 2808 * @len: bytes to move 2809 * 2810 * Increase the headroom of an empty &sk_buff by reducing the tail 2811 * room. This is only allowed for an empty buffer. 2812 */ 2813 static inline void skb_reserve(struct sk_buff *skb, int len) 2814 { 2815 skb->data += len; 2816 skb->tail += len; 2817 } 2818 2819 /** 2820 * skb_tailroom_reserve - adjust reserved_tailroom 2821 * @skb: buffer to alter 2822 * @mtu: maximum amount of headlen permitted 2823 * @needed_tailroom: minimum amount of reserved_tailroom 2824 * 2825 * Set reserved_tailroom so that headlen can be as large as possible but 2826 * not larger than mtu and tailroom cannot be smaller than 2827 * needed_tailroom. 2828 * The required headroom should already have been reserved before using 2829 * this function. 2830 */ 2831 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu, 2832 unsigned int needed_tailroom) 2833 { 2834 SKB_LINEAR_ASSERT(skb); 2835 if (mtu < skb_tailroom(skb) - needed_tailroom) 2836 /* use at most mtu */ 2837 skb->reserved_tailroom = skb_tailroom(skb) - mtu; 2838 else 2839 /* use up to all available space */ 2840 skb->reserved_tailroom = needed_tailroom; 2841 } 2842 2843 #define ENCAP_TYPE_ETHER 0 2844 #define ENCAP_TYPE_IPPROTO 1 2845 2846 static inline void skb_set_inner_protocol(struct sk_buff *skb, 2847 __be16 protocol) 2848 { 2849 skb->inner_protocol = protocol; 2850 skb->inner_protocol_type = ENCAP_TYPE_ETHER; 2851 } 2852 2853 static inline void skb_set_inner_ipproto(struct sk_buff *skb, 2854 __u8 ipproto) 2855 { 2856 skb->inner_ipproto = ipproto; 2857 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO; 2858 } 2859 2860 static inline void skb_reset_inner_headers(struct sk_buff *skb) 2861 { 2862 skb->inner_mac_header = skb->mac_header; 2863 skb->inner_network_header = skb->network_header; 2864 skb->inner_transport_header = skb->transport_header; 2865 } 2866 2867 static inline void skb_reset_mac_len(struct sk_buff *skb) 2868 { 2869 skb->mac_len = skb->network_header - skb->mac_header; 2870 } 2871 2872 static inline unsigned char *skb_inner_transport_header(const struct sk_buff 2873 *skb) 2874 { 2875 return skb->head + skb->inner_transport_header; 2876 } 2877 2878 static inline int skb_inner_transport_offset(const struct sk_buff *skb) 2879 { 2880 return skb_inner_transport_header(skb) - skb->data; 2881 } 2882 2883 static inline void skb_reset_inner_transport_header(struct sk_buff *skb) 2884 { 2885 skb->inner_transport_header = skb->data - skb->head; 2886 } 2887 2888 static inline void skb_set_inner_transport_header(struct sk_buff *skb, 2889 const int offset) 2890 { 2891 skb_reset_inner_transport_header(skb); 2892 skb->inner_transport_header += offset; 2893 } 2894 2895 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb) 2896 { 2897 return skb->head + skb->inner_network_header; 2898 } 2899 2900 static inline void skb_reset_inner_network_header(struct sk_buff *skb) 2901 { 2902 skb->inner_network_header = skb->data - skb->head; 2903 } 2904 2905 static inline void skb_set_inner_network_header(struct sk_buff *skb, 2906 const int offset) 2907 { 2908 skb_reset_inner_network_header(skb); 2909 skb->inner_network_header += offset; 2910 } 2911 2912 static inline bool skb_inner_network_header_was_set(const struct sk_buff *skb) 2913 { 2914 return skb->inner_network_header > 0; 2915 } 2916 2917 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb) 2918 { 2919 return skb->head + skb->inner_mac_header; 2920 } 2921 2922 static inline void skb_reset_inner_mac_header(struct sk_buff *skb) 2923 { 2924 skb->inner_mac_header = skb->data - skb->head; 2925 } 2926 2927 static inline void skb_set_inner_mac_header(struct sk_buff *skb, 2928 const int offset) 2929 { 2930 skb_reset_inner_mac_header(skb); 2931 skb->inner_mac_header += offset; 2932 } 2933 static inline bool skb_transport_header_was_set(const struct sk_buff *skb) 2934 { 2935 return skb->transport_header != (typeof(skb->transport_header))~0U; 2936 } 2937 2938 static inline unsigned char *skb_transport_header(const struct sk_buff *skb) 2939 { 2940 DEBUG_NET_WARN_ON_ONCE(!skb_transport_header_was_set(skb)); 2941 return skb->head + skb->transport_header; 2942 } 2943 2944 static inline void skb_reset_transport_header(struct sk_buff *skb) 2945 { 2946 skb->transport_header = skb->data - skb->head; 2947 } 2948 2949 static inline void skb_set_transport_header(struct sk_buff *skb, 2950 const int offset) 2951 { 2952 skb_reset_transport_header(skb); 2953 skb->transport_header += offset; 2954 } 2955 2956 static inline unsigned char *skb_network_header(const struct sk_buff *skb) 2957 { 2958 return skb->head + skb->network_header; 2959 } 2960 2961 static inline void skb_reset_network_header(struct sk_buff *skb) 2962 { 2963 skb->network_header = skb->data - skb->head; 2964 } 2965 2966 static inline void skb_set_network_header(struct sk_buff *skb, const int offset) 2967 { 2968 skb_reset_network_header(skb); 2969 skb->network_header += offset; 2970 } 2971 2972 static inline int skb_mac_header_was_set(const struct sk_buff *skb) 2973 { 2974 return skb->mac_header != (typeof(skb->mac_header))~0U; 2975 } 2976 2977 static inline unsigned char *skb_mac_header(const struct sk_buff *skb) 2978 { 2979 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb)); 2980 return skb->head + skb->mac_header; 2981 } 2982 2983 static inline int skb_mac_offset(const struct sk_buff *skb) 2984 { 2985 return skb_mac_header(skb) - skb->data; 2986 } 2987 2988 static inline u32 skb_mac_header_len(const struct sk_buff *skb) 2989 { 2990 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb)); 2991 return skb->network_header - skb->mac_header; 2992 } 2993 2994 static inline void skb_unset_mac_header(struct sk_buff *skb) 2995 { 2996 skb->mac_header = (typeof(skb->mac_header))~0U; 2997 } 2998 2999 static inline void skb_reset_mac_header(struct sk_buff *skb) 3000 { 3001 skb->mac_header = skb->data - skb->head; 3002 } 3003 3004 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset) 3005 { 3006 skb_reset_mac_header(skb); 3007 skb->mac_header += offset; 3008 } 3009 3010 static inline void skb_pop_mac_header(struct sk_buff *skb) 3011 { 3012 skb->mac_header = skb->network_header; 3013 } 3014 3015 static inline void skb_probe_transport_header(struct sk_buff *skb) 3016 { 3017 struct flow_keys_basic keys; 3018 3019 if (skb_transport_header_was_set(skb)) 3020 return; 3021 3022 if (skb_flow_dissect_flow_keys_basic(NULL, skb, &keys, 3023 NULL, 0, 0, 0, 0)) 3024 skb_set_transport_header(skb, keys.control.thoff); 3025 } 3026 3027 static inline void skb_mac_header_rebuild(struct sk_buff *skb) 3028 { 3029 if (skb_mac_header_was_set(skb)) { 3030 const unsigned char *old_mac = skb_mac_header(skb); 3031 3032 skb_set_mac_header(skb, -skb->mac_len); 3033 memmove(skb_mac_header(skb), old_mac, skb->mac_len); 3034 } 3035 } 3036 3037 /* Move the full mac header up to current network_header. 3038 * Leaves skb->data pointing at offset skb->mac_len into the mac_header. 3039 * Must be provided the complete mac header length. 3040 */ 3041 static inline void skb_mac_header_rebuild_full(struct sk_buff *skb, u32 full_mac_len) 3042 { 3043 if (skb_mac_header_was_set(skb)) { 3044 const unsigned char *old_mac = skb_mac_header(skb); 3045 3046 skb_set_mac_header(skb, -full_mac_len); 3047 memmove(skb_mac_header(skb), old_mac, full_mac_len); 3048 __skb_push(skb, full_mac_len - skb->mac_len); 3049 } 3050 } 3051 3052 static inline int skb_checksum_start_offset(const struct sk_buff *skb) 3053 { 3054 return skb->csum_start - skb_headroom(skb); 3055 } 3056 3057 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb) 3058 { 3059 return skb->head + skb->csum_start; 3060 } 3061 3062 static inline int skb_transport_offset(const struct sk_buff *skb) 3063 { 3064 return skb_transport_header(skb) - skb->data; 3065 } 3066 3067 static inline u32 skb_network_header_len(const struct sk_buff *skb) 3068 { 3069 DEBUG_NET_WARN_ON_ONCE(!skb_transport_header_was_set(skb)); 3070 return skb->transport_header - skb->network_header; 3071 } 3072 3073 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb) 3074 { 3075 return skb->inner_transport_header - skb->inner_network_header; 3076 } 3077 3078 static inline int skb_network_offset(const struct sk_buff *skb) 3079 { 3080 return skb_network_header(skb) - skb->data; 3081 } 3082 3083 static inline int skb_inner_network_offset(const struct sk_buff *skb) 3084 { 3085 return skb_inner_network_header(skb) - skb->data; 3086 } 3087 3088 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len) 3089 { 3090 return pskb_may_pull(skb, skb_network_offset(skb) + len); 3091 } 3092 3093 /* 3094 * CPUs often take a performance hit when accessing unaligned memory 3095 * locations. The actual performance hit varies, it can be small if the 3096 * hardware handles it or large if we have to take an exception and fix it 3097 * in software. 3098 * 3099 * Since an ethernet header is 14 bytes network drivers often end up with 3100 * the IP header at an unaligned offset. The IP header can be aligned by 3101 * shifting the start of the packet by 2 bytes. Drivers should do this 3102 * with: 3103 * 3104 * skb_reserve(skb, NET_IP_ALIGN); 3105 * 3106 * The downside to this alignment of the IP header is that the DMA is now 3107 * unaligned. On some architectures the cost of an unaligned DMA is high 3108 * and this cost outweighs the gains made by aligning the IP header. 3109 * 3110 * Since this trade off varies between architectures, we allow NET_IP_ALIGN 3111 * to be overridden. 3112 */ 3113 #ifndef NET_IP_ALIGN 3114 #define NET_IP_ALIGN 2 3115 #endif 3116 3117 /* 3118 * The networking layer reserves some headroom in skb data (via 3119 * dev_alloc_skb). This is used to avoid having to reallocate skb data when 3120 * the header has to grow. In the default case, if the header has to grow 3121 * 32 bytes or less we avoid the reallocation. 3122 * 3123 * Unfortunately this headroom changes the DMA alignment of the resulting 3124 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive 3125 * on some architectures. An architecture can override this value, 3126 * perhaps setting it to a cacheline in size (since that will maintain 3127 * cacheline alignment of the DMA). It must be a power of 2. 3128 * 3129 * Various parts of the networking layer expect at least 32 bytes of 3130 * headroom, you should not reduce this. 3131 * 3132 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS) 3133 * to reduce average number of cache lines per packet. 3134 * get_rps_cpu() for example only access one 64 bytes aligned block : 3135 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8) 3136 */ 3137 #ifndef NET_SKB_PAD 3138 #define NET_SKB_PAD max(32, L1_CACHE_BYTES) 3139 #endif 3140 3141 int ___pskb_trim(struct sk_buff *skb, unsigned int len); 3142 3143 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len) 3144 { 3145 if (WARN_ON(skb_is_nonlinear(skb))) 3146 return; 3147 skb->len = len; 3148 skb_set_tail_pointer(skb, len); 3149 } 3150 3151 static inline void __skb_trim(struct sk_buff *skb, unsigned int len) 3152 { 3153 __skb_set_length(skb, len); 3154 } 3155 3156 void skb_trim(struct sk_buff *skb, unsigned int len); 3157 3158 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len) 3159 { 3160 if (skb->data_len) 3161 return ___pskb_trim(skb, len); 3162 __skb_trim(skb, len); 3163 return 0; 3164 } 3165 3166 static inline int pskb_trim(struct sk_buff *skb, unsigned int len) 3167 { 3168 return (len < skb->len) ? __pskb_trim(skb, len) : 0; 3169 } 3170 3171 /** 3172 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer 3173 * @skb: buffer to alter 3174 * @len: new length 3175 * 3176 * This is identical to pskb_trim except that the caller knows that 3177 * the skb is not cloned so we should never get an error due to out- 3178 * of-memory. 3179 */ 3180 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len) 3181 { 3182 int err = pskb_trim(skb, len); 3183 BUG_ON(err); 3184 } 3185 3186 static inline int __skb_grow(struct sk_buff *skb, unsigned int len) 3187 { 3188 unsigned int diff = len - skb->len; 3189 3190 if (skb_tailroom(skb) < diff) { 3191 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb), 3192 GFP_ATOMIC); 3193 if (ret) 3194 return ret; 3195 } 3196 __skb_set_length(skb, len); 3197 return 0; 3198 } 3199 3200 /** 3201 * skb_orphan - orphan a buffer 3202 * @skb: buffer to orphan 3203 * 3204 * If a buffer currently has an owner then we call the owner's 3205 * destructor function and make the @skb unowned. The buffer continues 3206 * to exist but is no longer charged to its former owner. 3207 */ 3208 static inline void skb_orphan(struct sk_buff *skb) 3209 { 3210 if (skb->destructor) { 3211 skb->destructor(skb); 3212 skb->destructor = NULL; 3213 skb->sk = NULL; 3214 } else { 3215 BUG_ON(skb->sk); 3216 } 3217 } 3218 3219 /** 3220 * skb_orphan_frags - orphan the frags contained in a buffer 3221 * @skb: buffer to orphan frags from 3222 * @gfp_mask: allocation mask for replacement pages 3223 * 3224 * For each frag in the SKB which needs a destructor (i.e. has an 3225 * owner) create a copy of that frag and release the original 3226 * page by calling the destructor. 3227 */ 3228 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask) 3229 { 3230 if (likely(!skb_zcopy(skb))) 3231 return 0; 3232 if (skb_shinfo(skb)->flags & SKBFL_DONT_ORPHAN) 3233 return 0; 3234 return skb_copy_ubufs(skb, gfp_mask); 3235 } 3236 3237 /* Frags must be orphaned, even if refcounted, if skb might loop to rx path */ 3238 static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask) 3239 { 3240 if (likely(!skb_zcopy(skb))) 3241 return 0; 3242 return skb_copy_ubufs(skb, gfp_mask); 3243 } 3244 3245 /** 3246 * __skb_queue_purge_reason - empty a list 3247 * @list: list to empty 3248 * @reason: drop reason 3249 * 3250 * Delete all buffers on an &sk_buff list. Each buffer is removed from 3251 * the list and one reference dropped. This function does not take the 3252 * list lock and the caller must hold the relevant locks to use it. 3253 */ 3254 static inline void __skb_queue_purge_reason(struct sk_buff_head *list, 3255 enum skb_drop_reason reason) 3256 { 3257 struct sk_buff *skb; 3258 3259 while ((skb = __skb_dequeue(list)) != NULL) 3260 kfree_skb_reason(skb, reason); 3261 } 3262 3263 static inline void __skb_queue_purge(struct sk_buff_head *list) 3264 { 3265 __skb_queue_purge_reason(list, SKB_DROP_REASON_QUEUE_PURGE); 3266 } 3267 3268 void skb_queue_purge_reason(struct sk_buff_head *list, 3269 enum skb_drop_reason reason); 3270 3271 static inline void skb_queue_purge(struct sk_buff_head *list) 3272 { 3273 skb_queue_purge_reason(list, SKB_DROP_REASON_QUEUE_PURGE); 3274 } 3275 3276 unsigned int skb_rbtree_purge(struct rb_root *root); 3277 void skb_errqueue_purge(struct sk_buff_head *list); 3278 3279 void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask); 3280 3281 /** 3282 * netdev_alloc_frag - allocate a page fragment 3283 * @fragsz: fragment size 3284 * 3285 * Allocates a frag from a page for receive buffer. 3286 * Uses GFP_ATOMIC allocations. 3287 */ 3288 static inline void *netdev_alloc_frag(unsigned int fragsz) 3289 { 3290 return __netdev_alloc_frag_align(fragsz, ~0u); 3291 } 3292 3293 static inline void *netdev_alloc_frag_align(unsigned int fragsz, 3294 unsigned int align) 3295 { 3296 WARN_ON_ONCE(!is_power_of_2(align)); 3297 return __netdev_alloc_frag_align(fragsz, -align); 3298 } 3299 3300 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length, 3301 gfp_t gfp_mask); 3302 3303 /** 3304 * netdev_alloc_skb - allocate an skbuff for rx on a specific device 3305 * @dev: network device to receive on 3306 * @length: length to allocate 3307 * 3308 * Allocate a new &sk_buff and assign it a usage count of one. The 3309 * buffer has unspecified headroom built in. Users should allocate 3310 * the headroom they think they need without accounting for the 3311 * built in space. The built in space is used for optimisations. 3312 * 3313 * %NULL is returned if there is no free memory. Although this function 3314 * allocates memory it can be called from an interrupt. 3315 */ 3316 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev, 3317 unsigned int length) 3318 { 3319 return __netdev_alloc_skb(dev, length, GFP_ATOMIC); 3320 } 3321 3322 /* legacy helper around __netdev_alloc_skb() */ 3323 static inline struct sk_buff *__dev_alloc_skb(unsigned int length, 3324 gfp_t gfp_mask) 3325 { 3326 return __netdev_alloc_skb(NULL, length, gfp_mask); 3327 } 3328 3329 /* legacy helper around netdev_alloc_skb() */ 3330 static inline struct sk_buff *dev_alloc_skb(unsigned int length) 3331 { 3332 return netdev_alloc_skb(NULL, length); 3333 } 3334 3335 3336 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev, 3337 unsigned int length, gfp_t gfp) 3338 { 3339 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp); 3340 3341 if (NET_IP_ALIGN && skb) 3342 skb_reserve(skb, NET_IP_ALIGN); 3343 return skb; 3344 } 3345 3346 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev, 3347 unsigned int length) 3348 { 3349 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC); 3350 } 3351 3352 static inline void skb_free_frag(void *addr) 3353 { 3354 page_frag_free(addr); 3355 } 3356 3357 void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask); 3358 3359 static inline void *napi_alloc_frag(unsigned int fragsz) 3360 { 3361 return __napi_alloc_frag_align(fragsz, ~0u); 3362 } 3363 3364 static inline void *napi_alloc_frag_align(unsigned int fragsz, 3365 unsigned int align) 3366 { 3367 WARN_ON_ONCE(!is_power_of_2(align)); 3368 return __napi_alloc_frag_align(fragsz, -align); 3369 } 3370 3371 struct sk_buff *napi_alloc_skb(struct napi_struct *napi, unsigned int length); 3372 void napi_consume_skb(struct sk_buff *skb, int budget); 3373 3374 void napi_skb_free_stolen_head(struct sk_buff *skb); 3375 void __napi_kfree_skb(struct sk_buff *skb, enum skb_drop_reason reason); 3376 3377 /** 3378 * __dev_alloc_pages - allocate page for network Rx 3379 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 3380 * @order: size of the allocation 3381 * 3382 * Allocate a new page. 3383 * 3384 * %NULL is returned if there is no free memory. 3385 */ 3386 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask, 3387 unsigned int order) 3388 { 3389 /* This piece of code contains several assumptions. 3390 * 1. This is for device Rx, therefore a cold page is preferred. 3391 * 2. The expectation is the user wants a compound page. 3392 * 3. If requesting a order 0 page it will not be compound 3393 * due to the check to see if order has a value in prep_new_page 3394 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to 3395 * code in gfp_to_alloc_flags that should be enforcing this. 3396 */ 3397 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC; 3398 3399 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order); 3400 } 3401 3402 static inline struct page *dev_alloc_pages(unsigned int order) 3403 { 3404 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order); 3405 } 3406 3407 /** 3408 * __dev_alloc_page - allocate a page for network Rx 3409 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 3410 * 3411 * Allocate a new page. 3412 * 3413 * %NULL is returned if there is no free memory. 3414 */ 3415 static inline struct page *__dev_alloc_page(gfp_t gfp_mask) 3416 { 3417 return __dev_alloc_pages(gfp_mask, 0); 3418 } 3419 3420 static inline struct page *dev_alloc_page(void) 3421 { 3422 return dev_alloc_pages(0); 3423 } 3424 3425 /** 3426 * dev_page_is_reusable - check whether a page can be reused for network Rx 3427 * @page: the page to test 3428 * 3429 * A page shouldn't be considered for reusing/recycling if it was allocated 3430 * under memory pressure or at a distant memory node. 3431 * 3432 * Returns false if this page should be returned to page allocator, true 3433 * otherwise. 3434 */ 3435 static inline bool dev_page_is_reusable(const struct page *page) 3436 { 3437 return likely(page_to_nid(page) == numa_mem_id() && 3438 !page_is_pfmemalloc(page)); 3439 } 3440 3441 /** 3442 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page 3443 * @page: The page that was allocated from skb_alloc_page 3444 * @skb: The skb that may need pfmemalloc set 3445 */ 3446 static inline void skb_propagate_pfmemalloc(const struct page *page, 3447 struct sk_buff *skb) 3448 { 3449 if (page_is_pfmemalloc(page)) 3450 skb->pfmemalloc = true; 3451 } 3452 3453 /** 3454 * skb_frag_off() - Returns the offset of a skb fragment 3455 * @frag: the paged fragment 3456 */ 3457 static inline unsigned int skb_frag_off(const skb_frag_t *frag) 3458 { 3459 return frag->offset; 3460 } 3461 3462 /** 3463 * skb_frag_off_add() - Increments the offset of a skb fragment by @delta 3464 * @frag: skb fragment 3465 * @delta: value to add 3466 */ 3467 static inline void skb_frag_off_add(skb_frag_t *frag, int delta) 3468 { 3469 frag->offset += delta; 3470 } 3471 3472 /** 3473 * skb_frag_off_set() - Sets the offset of a skb fragment 3474 * @frag: skb fragment 3475 * @offset: offset of fragment 3476 */ 3477 static inline void skb_frag_off_set(skb_frag_t *frag, unsigned int offset) 3478 { 3479 frag->offset = offset; 3480 } 3481 3482 /** 3483 * skb_frag_off_copy() - Sets the offset of a skb fragment from another fragment 3484 * @fragto: skb fragment where offset is set 3485 * @fragfrom: skb fragment offset is copied from 3486 */ 3487 static inline void skb_frag_off_copy(skb_frag_t *fragto, 3488 const skb_frag_t *fragfrom) 3489 { 3490 fragto->offset = fragfrom->offset; 3491 } 3492 3493 /** 3494 * skb_frag_page - retrieve the page referred to by a paged fragment 3495 * @frag: the paged fragment 3496 * 3497 * Returns the &struct page associated with @frag. 3498 */ 3499 static inline struct page *skb_frag_page(const skb_frag_t *frag) 3500 { 3501 return netmem_to_page(frag->netmem); 3502 } 3503 3504 int skb_pp_cow_data(struct page_pool *pool, struct sk_buff **pskb, 3505 unsigned int headroom); 3506 int skb_cow_data_for_xdp(struct page_pool *pool, struct sk_buff **pskb, 3507 struct bpf_prog *prog); 3508 /** 3509 * skb_frag_address - gets the address of the data contained in a paged fragment 3510 * @frag: the paged fragment buffer 3511 * 3512 * Returns the address of the data within @frag. The page must already 3513 * be mapped. 3514 */ 3515 static inline void *skb_frag_address(const skb_frag_t *frag) 3516 { 3517 return page_address(skb_frag_page(frag)) + skb_frag_off(frag); 3518 } 3519 3520 /** 3521 * skb_frag_address_safe - gets the address of the data contained in a paged fragment 3522 * @frag: the paged fragment buffer 3523 * 3524 * Returns the address of the data within @frag. Checks that the page 3525 * is mapped and returns %NULL otherwise. 3526 */ 3527 static inline void *skb_frag_address_safe(const skb_frag_t *frag) 3528 { 3529 void *ptr = page_address(skb_frag_page(frag)); 3530 if (unlikely(!ptr)) 3531 return NULL; 3532 3533 return ptr + skb_frag_off(frag); 3534 } 3535 3536 /** 3537 * skb_frag_page_copy() - sets the page in a fragment from another fragment 3538 * @fragto: skb fragment where page is set 3539 * @fragfrom: skb fragment page is copied from 3540 */ 3541 static inline void skb_frag_page_copy(skb_frag_t *fragto, 3542 const skb_frag_t *fragfrom) 3543 { 3544 fragto->netmem = fragfrom->netmem; 3545 } 3546 3547 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio); 3548 3549 /** 3550 * skb_frag_dma_map - maps a paged fragment via the DMA API 3551 * @dev: the device to map the fragment to 3552 * @frag: the paged fragment to map 3553 * @offset: the offset within the fragment (starting at the 3554 * fragment's own offset) 3555 * @size: the number of bytes to map 3556 * @dir: the direction of the mapping (``PCI_DMA_*``) 3557 * 3558 * Maps the page associated with @frag to @device. 3559 */ 3560 static inline dma_addr_t skb_frag_dma_map(struct device *dev, 3561 const skb_frag_t *frag, 3562 size_t offset, size_t size, 3563 enum dma_data_direction dir) 3564 { 3565 return dma_map_page(dev, skb_frag_page(frag), 3566 skb_frag_off(frag) + offset, size, dir); 3567 } 3568 3569 static inline struct sk_buff *pskb_copy(struct sk_buff *skb, 3570 gfp_t gfp_mask) 3571 { 3572 return __pskb_copy(skb, skb_headroom(skb), gfp_mask); 3573 } 3574 3575 3576 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb, 3577 gfp_t gfp_mask) 3578 { 3579 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true); 3580 } 3581 3582 3583 /** 3584 * skb_clone_writable - is the header of a clone writable 3585 * @skb: buffer to check 3586 * @len: length up to which to write 3587 * 3588 * Returns true if modifying the header part of the cloned buffer 3589 * does not requires the data to be copied. 3590 */ 3591 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len) 3592 { 3593 return !skb_header_cloned(skb) && 3594 skb_headroom(skb) + len <= skb->hdr_len; 3595 } 3596 3597 static inline int skb_try_make_writable(struct sk_buff *skb, 3598 unsigned int write_len) 3599 { 3600 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) && 3601 pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 3602 } 3603 3604 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom, 3605 int cloned) 3606 { 3607 int delta = 0; 3608 3609 if (headroom > skb_headroom(skb)) 3610 delta = headroom - skb_headroom(skb); 3611 3612 if (delta || cloned) 3613 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0, 3614 GFP_ATOMIC); 3615 return 0; 3616 } 3617 3618 /** 3619 * skb_cow - copy header of skb when it is required 3620 * @skb: buffer to cow 3621 * @headroom: needed headroom 3622 * 3623 * If the skb passed lacks sufficient headroom or its data part 3624 * is shared, data is reallocated. If reallocation fails, an error 3625 * is returned and original skb is not changed. 3626 * 3627 * The result is skb with writable area skb->head...skb->tail 3628 * and at least @headroom of space at head. 3629 */ 3630 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom) 3631 { 3632 return __skb_cow(skb, headroom, skb_cloned(skb)); 3633 } 3634 3635 /** 3636 * skb_cow_head - skb_cow but only making the head writable 3637 * @skb: buffer to cow 3638 * @headroom: needed headroom 3639 * 3640 * This function is identical to skb_cow except that we replace the 3641 * skb_cloned check by skb_header_cloned. It should be used when 3642 * you only need to push on some header and do not need to modify 3643 * the data. 3644 */ 3645 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom) 3646 { 3647 return __skb_cow(skb, headroom, skb_header_cloned(skb)); 3648 } 3649 3650 /** 3651 * skb_padto - pad an skbuff up to a minimal size 3652 * @skb: buffer to pad 3653 * @len: minimal length 3654 * 3655 * Pads up a buffer to ensure the trailing bytes exist and are 3656 * blanked. If the buffer already contains sufficient data it 3657 * is untouched. Otherwise it is extended. Returns zero on 3658 * success. The skb is freed on error. 3659 */ 3660 static inline int skb_padto(struct sk_buff *skb, unsigned int len) 3661 { 3662 unsigned int size = skb->len; 3663 if (likely(size >= len)) 3664 return 0; 3665 return skb_pad(skb, len - size); 3666 } 3667 3668 /** 3669 * __skb_put_padto - increase size and pad an skbuff up to a minimal size 3670 * @skb: buffer to pad 3671 * @len: minimal length 3672 * @free_on_error: free buffer on error 3673 * 3674 * Pads up a buffer to ensure the trailing bytes exist and are 3675 * blanked. If the buffer already contains sufficient data it 3676 * is untouched. Otherwise it is extended. Returns zero on 3677 * success. The skb is freed on error if @free_on_error is true. 3678 */ 3679 static inline int __must_check __skb_put_padto(struct sk_buff *skb, 3680 unsigned int len, 3681 bool free_on_error) 3682 { 3683 unsigned int size = skb->len; 3684 3685 if (unlikely(size < len)) { 3686 len -= size; 3687 if (__skb_pad(skb, len, free_on_error)) 3688 return -ENOMEM; 3689 __skb_put(skb, len); 3690 } 3691 return 0; 3692 } 3693 3694 /** 3695 * skb_put_padto - increase size and pad an skbuff up to a minimal size 3696 * @skb: buffer to pad 3697 * @len: minimal length 3698 * 3699 * Pads up a buffer to ensure the trailing bytes exist and are 3700 * blanked. If the buffer already contains sufficient data it 3701 * is untouched. Otherwise it is extended. Returns zero on 3702 * success. The skb is freed on error. 3703 */ 3704 static inline int __must_check skb_put_padto(struct sk_buff *skb, unsigned int len) 3705 { 3706 return __skb_put_padto(skb, len, true); 3707 } 3708 3709 bool csum_and_copy_from_iter_full(void *addr, size_t bytes, __wsum *csum, struct iov_iter *i) 3710 __must_check; 3711 3712 static inline int skb_add_data(struct sk_buff *skb, 3713 struct iov_iter *from, int copy) 3714 { 3715 const int off = skb->len; 3716 3717 if (skb->ip_summed == CHECKSUM_NONE) { 3718 __wsum csum = 0; 3719 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy, 3720 &csum, from)) { 3721 skb->csum = csum_block_add(skb->csum, csum, off); 3722 return 0; 3723 } 3724 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from)) 3725 return 0; 3726 3727 __skb_trim(skb, off); 3728 return -EFAULT; 3729 } 3730 3731 static inline bool skb_can_coalesce(struct sk_buff *skb, int i, 3732 const struct page *page, int off) 3733 { 3734 if (skb_zcopy(skb)) 3735 return false; 3736 if (i) { 3737 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1]; 3738 3739 return page == skb_frag_page(frag) && 3740 off == skb_frag_off(frag) + skb_frag_size(frag); 3741 } 3742 return false; 3743 } 3744 3745 static inline int __skb_linearize(struct sk_buff *skb) 3746 { 3747 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM; 3748 } 3749 3750 /** 3751 * skb_linearize - convert paged skb to linear one 3752 * @skb: buffer to linarize 3753 * 3754 * If there is no free memory -ENOMEM is returned, otherwise zero 3755 * is returned and the old skb data released. 3756 */ 3757 static inline int skb_linearize(struct sk_buff *skb) 3758 { 3759 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0; 3760 } 3761 3762 /** 3763 * skb_has_shared_frag - can any frag be overwritten 3764 * @skb: buffer to test 3765 * 3766 * Return true if the skb has at least one frag that might be modified 3767 * by an external entity (as in vmsplice()/sendfile()) 3768 */ 3769 static inline bool skb_has_shared_frag(const struct sk_buff *skb) 3770 { 3771 return skb_is_nonlinear(skb) && 3772 skb_shinfo(skb)->flags & SKBFL_SHARED_FRAG; 3773 } 3774 3775 /** 3776 * skb_linearize_cow - make sure skb is linear and writable 3777 * @skb: buffer to process 3778 * 3779 * If there is no free memory -ENOMEM is returned, otherwise zero 3780 * is returned and the old skb data released. 3781 */ 3782 static inline int skb_linearize_cow(struct sk_buff *skb) 3783 { 3784 return skb_is_nonlinear(skb) || skb_cloned(skb) ? 3785 __skb_linearize(skb) : 0; 3786 } 3787 3788 static __always_inline void 3789 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 3790 unsigned int off) 3791 { 3792 if (skb->ip_summed == CHECKSUM_COMPLETE) 3793 skb->csum = csum_block_sub(skb->csum, 3794 csum_partial(start, len, 0), off); 3795 else if (skb->ip_summed == CHECKSUM_PARTIAL && 3796 skb_checksum_start_offset(skb) < 0) 3797 skb->ip_summed = CHECKSUM_NONE; 3798 } 3799 3800 /** 3801 * skb_postpull_rcsum - update checksum for received skb after pull 3802 * @skb: buffer to update 3803 * @start: start of data before pull 3804 * @len: length of data pulled 3805 * 3806 * After doing a pull on a received packet, you need to call this to 3807 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to 3808 * CHECKSUM_NONE so that it can be recomputed from scratch. 3809 */ 3810 static inline void skb_postpull_rcsum(struct sk_buff *skb, 3811 const void *start, unsigned int len) 3812 { 3813 if (skb->ip_summed == CHECKSUM_COMPLETE) 3814 skb->csum = wsum_negate(csum_partial(start, len, 3815 wsum_negate(skb->csum))); 3816 else if (skb->ip_summed == CHECKSUM_PARTIAL && 3817 skb_checksum_start_offset(skb) < 0) 3818 skb->ip_summed = CHECKSUM_NONE; 3819 } 3820 3821 static __always_inline void 3822 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 3823 unsigned int off) 3824 { 3825 if (skb->ip_summed == CHECKSUM_COMPLETE) 3826 skb->csum = csum_block_add(skb->csum, 3827 csum_partial(start, len, 0), off); 3828 } 3829 3830 /** 3831 * skb_postpush_rcsum - update checksum for received skb after push 3832 * @skb: buffer to update 3833 * @start: start of data after push 3834 * @len: length of data pushed 3835 * 3836 * After doing a push on a received packet, you need to call this to 3837 * update the CHECKSUM_COMPLETE checksum. 3838 */ 3839 static inline void skb_postpush_rcsum(struct sk_buff *skb, 3840 const void *start, unsigned int len) 3841 { 3842 __skb_postpush_rcsum(skb, start, len, 0); 3843 } 3844 3845 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len); 3846 3847 /** 3848 * skb_push_rcsum - push skb and update receive checksum 3849 * @skb: buffer to update 3850 * @len: length of data pulled 3851 * 3852 * This function performs an skb_push on the packet and updates 3853 * the CHECKSUM_COMPLETE checksum. It should be used on 3854 * receive path processing instead of skb_push unless you know 3855 * that the checksum difference is zero (e.g., a valid IP header) 3856 * or you are setting ip_summed to CHECKSUM_NONE. 3857 */ 3858 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len) 3859 { 3860 skb_push(skb, len); 3861 skb_postpush_rcsum(skb, skb->data, len); 3862 return skb->data; 3863 } 3864 3865 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len); 3866 /** 3867 * pskb_trim_rcsum - trim received skb and update checksum 3868 * @skb: buffer to trim 3869 * @len: new length 3870 * 3871 * This is exactly the same as pskb_trim except that it ensures the 3872 * checksum of received packets are still valid after the operation. 3873 * It can change skb pointers. 3874 */ 3875 3876 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len) 3877 { 3878 if (likely(len >= skb->len)) 3879 return 0; 3880 return pskb_trim_rcsum_slow(skb, len); 3881 } 3882 3883 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len) 3884 { 3885 if (skb->ip_summed == CHECKSUM_COMPLETE) 3886 skb->ip_summed = CHECKSUM_NONE; 3887 __skb_trim(skb, len); 3888 return 0; 3889 } 3890 3891 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len) 3892 { 3893 if (skb->ip_summed == CHECKSUM_COMPLETE) 3894 skb->ip_summed = CHECKSUM_NONE; 3895 return __skb_grow(skb, len); 3896 } 3897 3898 #define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode) 3899 #define skb_rb_first(root) rb_to_skb(rb_first(root)) 3900 #define skb_rb_last(root) rb_to_skb(rb_last(root)) 3901 #define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode)) 3902 #define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode)) 3903 3904 #define skb_queue_walk(queue, skb) \ 3905 for (skb = (queue)->next; \ 3906 skb != (struct sk_buff *)(queue); \ 3907 skb = skb->next) 3908 3909 #define skb_queue_walk_safe(queue, skb, tmp) \ 3910 for (skb = (queue)->next, tmp = skb->next; \ 3911 skb != (struct sk_buff *)(queue); \ 3912 skb = tmp, tmp = skb->next) 3913 3914 #define skb_queue_walk_from(queue, skb) \ 3915 for (; skb != (struct sk_buff *)(queue); \ 3916 skb = skb->next) 3917 3918 #define skb_rbtree_walk(skb, root) \ 3919 for (skb = skb_rb_first(root); skb != NULL; \ 3920 skb = skb_rb_next(skb)) 3921 3922 #define skb_rbtree_walk_from(skb) \ 3923 for (; skb != NULL; \ 3924 skb = skb_rb_next(skb)) 3925 3926 #define skb_rbtree_walk_from_safe(skb, tmp) \ 3927 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \ 3928 skb = tmp) 3929 3930 #define skb_queue_walk_from_safe(queue, skb, tmp) \ 3931 for (tmp = skb->next; \ 3932 skb != (struct sk_buff *)(queue); \ 3933 skb = tmp, tmp = skb->next) 3934 3935 #define skb_queue_reverse_walk(queue, skb) \ 3936 for (skb = (queue)->prev; \ 3937 skb != (struct sk_buff *)(queue); \ 3938 skb = skb->prev) 3939 3940 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \ 3941 for (skb = (queue)->prev, tmp = skb->prev; \ 3942 skb != (struct sk_buff *)(queue); \ 3943 skb = tmp, tmp = skb->prev) 3944 3945 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \ 3946 for (tmp = skb->prev; \ 3947 skb != (struct sk_buff *)(queue); \ 3948 skb = tmp, tmp = skb->prev) 3949 3950 static inline bool skb_has_frag_list(const struct sk_buff *skb) 3951 { 3952 return skb_shinfo(skb)->frag_list != NULL; 3953 } 3954 3955 static inline void skb_frag_list_init(struct sk_buff *skb) 3956 { 3957 skb_shinfo(skb)->frag_list = NULL; 3958 } 3959 3960 #define skb_walk_frags(skb, iter) \ 3961 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next) 3962 3963 3964 int __skb_wait_for_more_packets(struct sock *sk, struct sk_buff_head *queue, 3965 int *err, long *timeo_p, 3966 const struct sk_buff *skb); 3967 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk, 3968 struct sk_buff_head *queue, 3969 unsigned int flags, 3970 int *off, int *err, 3971 struct sk_buff **last); 3972 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, 3973 struct sk_buff_head *queue, 3974 unsigned int flags, int *off, int *err, 3975 struct sk_buff **last); 3976 struct sk_buff *__skb_recv_datagram(struct sock *sk, 3977 struct sk_buff_head *sk_queue, 3978 unsigned int flags, int *off, int *err); 3979 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags, int *err); 3980 __poll_t datagram_poll(struct file *file, struct socket *sock, 3981 struct poll_table_struct *wait); 3982 int skb_copy_datagram_iter(const struct sk_buff *from, int offset, 3983 struct iov_iter *to, int size); 3984 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset, 3985 struct msghdr *msg, int size) 3986 { 3987 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size); 3988 } 3989 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen, 3990 struct msghdr *msg); 3991 int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset, 3992 struct iov_iter *to, int len, 3993 struct ahash_request *hash); 3994 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset, 3995 struct iov_iter *from, int len); 3996 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm); 3997 void skb_free_datagram(struct sock *sk, struct sk_buff *skb); 3998 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags); 3999 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len); 4000 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len); 4001 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to, 4002 int len); 4003 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 4004 struct pipe_inode_info *pipe, unsigned int len, 4005 unsigned int flags); 4006 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset, 4007 int len); 4008 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len); 4009 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to); 4010 unsigned int skb_zerocopy_headlen(const struct sk_buff *from); 4011 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from, 4012 int len, int hlen); 4013 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len); 4014 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen); 4015 void skb_scrub_packet(struct sk_buff *skb, bool xnet); 4016 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features); 4017 struct sk_buff *skb_segment_list(struct sk_buff *skb, netdev_features_t features, 4018 unsigned int offset); 4019 struct sk_buff *skb_vlan_untag(struct sk_buff *skb); 4020 int skb_ensure_writable(struct sk_buff *skb, unsigned int write_len); 4021 int skb_ensure_writable_head_tail(struct sk_buff *skb, struct net_device *dev); 4022 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci); 4023 int skb_vlan_pop(struct sk_buff *skb); 4024 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci); 4025 int skb_eth_pop(struct sk_buff *skb); 4026 int skb_eth_push(struct sk_buff *skb, const unsigned char *dst, 4027 const unsigned char *src); 4028 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto, 4029 int mac_len, bool ethernet); 4030 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len, 4031 bool ethernet); 4032 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse); 4033 int skb_mpls_dec_ttl(struct sk_buff *skb); 4034 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy, 4035 gfp_t gfp); 4036 4037 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len) 4038 { 4039 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT; 4040 } 4041 4042 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len) 4043 { 4044 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 4045 } 4046 4047 struct skb_checksum_ops { 4048 __wsum (*update)(const void *mem, int len, __wsum wsum); 4049 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len); 4050 }; 4051 4052 extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly; 4053 4054 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 4055 __wsum csum, const struct skb_checksum_ops *ops); 4056 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len, 4057 __wsum csum); 4058 4059 static inline void * __must_check 4060 __skb_header_pointer(const struct sk_buff *skb, int offset, int len, 4061 const void *data, int hlen, void *buffer) 4062 { 4063 if (likely(hlen - offset >= len)) 4064 return (void *)data + offset; 4065 4066 if (!skb || unlikely(skb_copy_bits(skb, offset, buffer, len) < 0)) 4067 return NULL; 4068 4069 return buffer; 4070 } 4071 4072 static inline void * __must_check 4073 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer) 4074 { 4075 return __skb_header_pointer(skb, offset, len, skb->data, 4076 skb_headlen(skb), buffer); 4077 } 4078 4079 static inline void * __must_check 4080 skb_pointer_if_linear(const struct sk_buff *skb, int offset, int len) 4081 { 4082 if (likely(skb_headlen(skb) - offset >= len)) 4083 return skb->data + offset; 4084 return NULL; 4085 } 4086 4087 /** 4088 * skb_needs_linearize - check if we need to linearize a given skb 4089 * depending on the given device features. 4090 * @skb: socket buffer to check 4091 * @features: net device features 4092 * 4093 * Returns true if either: 4094 * 1. skb has frag_list and the device doesn't support FRAGLIST, or 4095 * 2. skb is fragmented and the device does not support SG. 4096 */ 4097 static inline bool skb_needs_linearize(struct sk_buff *skb, 4098 netdev_features_t features) 4099 { 4100 return skb_is_nonlinear(skb) && 4101 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) || 4102 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG))); 4103 } 4104 4105 static inline void skb_copy_from_linear_data(const struct sk_buff *skb, 4106 void *to, 4107 const unsigned int len) 4108 { 4109 memcpy(to, skb->data, len); 4110 } 4111 4112 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb, 4113 const int offset, void *to, 4114 const unsigned int len) 4115 { 4116 memcpy(to, skb->data + offset, len); 4117 } 4118 4119 static inline void skb_copy_to_linear_data(struct sk_buff *skb, 4120 const void *from, 4121 const unsigned int len) 4122 { 4123 memcpy(skb->data, from, len); 4124 } 4125 4126 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb, 4127 const int offset, 4128 const void *from, 4129 const unsigned int len) 4130 { 4131 memcpy(skb->data + offset, from, len); 4132 } 4133 4134 void skb_init(void); 4135 4136 static inline ktime_t skb_get_ktime(const struct sk_buff *skb) 4137 { 4138 return skb->tstamp; 4139 } 4140 4141 /** 4142 * skb_get_timestamp - get timestamp from a skb 4143 * @skb: skb to get stamp from 4144 * @stamp: pointer to struct __kernel_old_timeval to store stamp in 4145 * 4146 * Timestamps are stored in the skb as offsets to a base timestamp. 4147 * This function converts the offset back to a struct timeval and stores 4148 * it in stamp. 4149 */ 4150 static inline void skb_get_timestamp(const struct sk_buff *skb, 4151 struct __kernel_old_timeval *stamp) 4152 { 4153 *stamp = ns_to_kernel_old_timeval(skb->tstamp); 4154 } 4155 4156 static inline void skb_get_new_timestamp(const struct sk_buff *skb, 4157 struct __kernel_sock_timeval *stamp) 4158 { 4159 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4160 4161 stamp->tv_sec = ts.tv_sec; 4162 stamp->tv_usec = ts.tv_nsec / 1000; 4163 } 4164 4165 static inline void skb_get_timestampns(const struct sk_buff *skb, 4166 struct __kernel_old_timespec *stamp) 4167 { 4168 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4169 4170 stamp->tv_sec = ts.tv_sec; 4171 stamp->tv_nsec = ts.tv_nsec; 4172 } 4173 4174 static inline void skb_get_new_timestampns(const struct sk_buff *skb, 4175 struct __kernel_timespec *stamp) 4176 { 4177 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4178 4179 stamp->tv_sec = ts.tv_sec; 4180 stamp->tv_nsec = ts.tv_nsec; 4181 } 4182 4183 static inline void __net_timestamp(struct sk_buff *skb) 4184 { 4185 skb->tstamp = ktime_get_real(); 4186 skb->mono_delivery_time = 0; 4187 } 4188 4189 static inline ktime_t net_timedelta(ktime_t t) 4190 { 4191 return ktime_sub(ktime_get_real(), t); 4192 } 4193 4194 static inline void skb_set_delivery_time(struct sk_buff *skb, ktime_t kt, 4195 bool mono) 4196 { 4197 skb->tstamp = kt; 4198 skb->mono_delivery_time = kt && mono; 4199 } 4200 4201 DECLARE_STATIC_KEY_FALSE(netstamp_needed_key); 4202 4203 /* It is used in the ingress path to clear the delivery_time. 4204 * If needed, set the skb->tstamp to the (rcv) timestamp. 4205 */ 4206 static inline void skb_clear_delivery_time(struct sk_buff *skb) 4207 { 4208 if (skb->mono_delivery_time) { 4209 skb->mono_delivery_time = 0; 4210 if (static_branch_unlikely(&netstamp_needed_key)) 4211 skb->tstamp = ktime_get_real(); 4212 else 4213 skb->tstamp = 0; 4214 } 4215 } 4216 4217 static inline void skb_clear_tstamp(struct sk_buff *skb) 4218 { 4219 if (skb->mono_delivery_time) 4220 return; 4221 4222 skb->tstamp = 0; 4223 } 4224 4225 static inline ktime_t skb_tstamp(const struct sk_buff *skb) 4226 { 4227 if (skb->mono_delivery_time) 4228 return 0; 4229 4230 return skb->tstamp; 4231 } 4232 4233 static inline ktime_t skb_tstamp_cond(const struct sk_buff *skb, bool cond) 4234 { 4235 if (!skb->mono_delivery_time && skb->tstamp) 4236 return skb->tstamp; 4237 4238 if (static_branch_unlikely(&netstamp_needed_key) || cond) 4239 return ktime_get_real(); 4240 4241 return 0; 4242 } 4243 4244 static inline u8 skb_metadata_len(const struct sk_buff *skb) 4245 { 4246 return skb_shinfo(skb)->meta_len; 4247 } 4248 4249 static inline void *skb_metadata_end(const struct sk_buff *skb) 4250 { 4251 return skb_mac_header(skb); 4252 } 4253 4254 static inline bool __skb_metadata_differs(const struct sk_buff *skb_a, 4255 const struct sk_buff *skb_b, 4256 u8 meta_len) 4257 { 4258 const void *a = skb_metadata_end(skb_a); 4259 const void *b = skb_metadata_end(skb_b); 4260 u64 diffs = 0; 4261 4262 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) || 4263 BITS_PER_LONG != 64) 4264 goto slow; 4265 4266 /* Using more efficient variant than plain call to memcmp(). */ 4267 switch (meta_len) { 4268 #define __it(x, op) (x -= sizeof(u##op)) 4269 #define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op)) 4270 case 32: diffs |= __it_diff(a, b, 64); 4271 fallthrough; 4272 case 24: diffs |= __it_diff(a, b, 64); 4273 fallthrough; 4274 case 16: diffs |= __it_diff(a, b, 64); 4275 fallthrough; 4276 case 8: diffs |= __it_diff(a, b, 64); 4277 break; 4278 case 28: diffs |= __it_diff(a, b, 64); 4279 fallthrough; 4280 case 20: diffs |= __it_diff(a, b, 64); 4281 fallthrough; 4282 case 12: diffs |= __it_diff(a, b, 64); 4283 fallthrough; 4284 case 4: diffs |= __it_diff(a, b, 32); 4285 break; 4286 default: 4287 slow: 4288 return memcmp(a - meta_len, b - meta_len, meta_len); 4289 } 4290 return diffs; 4291 } 4292 4293 static inline bool skb_metadata_differs(const struct sk_buff *skb_a, 4294 const struct sk_buff *skb_b) 4295 { 4296 u8 len_a = skb_metadata_len(skb_a); 4297 u8 len_b = skb_metadata_len(skb_b); 4298 4299 if (!(len_a | len_b)) 4300 return false; 4301 4302 return len_a != len_b ? 4303 true : __skb_metadata_differs(skb_a, skb_b, len_a); 4304 } 4305 4306 static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len) 4307 { 4308 skb_shinfo(skb)->meta_len = meta_len; 4309 } 4310 4311 static inline void skb_metadata_clear(struct sk_buff *skb) 4312 { 4313 skb_metadata_set(skb, 0); 4314 } 4315 4316 struct sk_buff *skb_clone_sk(struct sk_buff *skb); 4317 4318 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING 4319 4320 void skb_clone_tx_timestamp(struct sk_buff *skb); 4321 bool skb_defer_rx_timestamp(struct sk_buff *skb); 4322 4323 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */ 4324 4325 static inline void skb_clone_tx_timestamp(struct sk_buff *skb) 4326 { 4327 } 4328 4329 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb) 4330 { 4331 return false; 4332 } 4333 4334 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */ 4335 4336 /** 4337 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps 4338 * 4339 * PHY drivers may accept clones of transmitted packets for 4340 * timestamping via their phy_driver.txtstamp method. These drivers 4341 * must call this function to return the skb back to the stack with a 4342 * timestamp. 4343 * 4344 * @skb: clone of the original outgoing packet 4345 * @hwtstamps: hardware time stamps 4346 * 4347 */ 4348 void skb_complete_tx_timestamp(struct sk_buff *skb, 4349 struct skb_shared_hwtstamps *hwtstamps); 4350 4351 void __skb_tstamp_tx(struct sk_buff *orig_skb, const struct sk_buff *ack_skb, 4352 struct skb_shared_hwtstamps *hwtstamps, 4353 struct sock *sk, int tstype); 4354 4355 /** 4356 * skb_tstamp_tx - queue clone of skb with send time stamps 4357 * @orig_skb: the original outgoing packet 4358 * @hwtstamps: hardware time stamps, may be NULL if not available 4359 * 4360 * If the skb has a socket associated, then this function clones the 4361 * skb (thus sharing the actual data and optional structures), stores 4362 * the optional hardware time stamping information (if non NULL) or 4363 * generates a software time stamp (otherwise), then queues the clone 4364 * to the error queue of the socket. Errors are silently ignored. 4365 */ 4366 void skb_tstamp_tx(struct sk_buff *orig_skb, 4367 struct skb_shared_hwtstamps *hwtstamps); 4368 4369 /** 4370 * skb_tx_timestamp() - Driver hook for transmit timestamping 4371 * 4372 * Ethernet MAC Drivers should call this function in their hard_xmit() 4373 * function immediately before giving the sk_buff to the MAC hardware. 4374 * 4375 * Specifically, one should make absolutely sure that this function is 4376 * called before TX completion of this packet can trigger. Otherwise 4377 * the packet could potentially already be freed. 4378 * 4379 * @skb: A socket buffer. 4380 */ 4381 static inline void skb_tx_timestamp(struct sk_buff *skb) 4382 { 4383 skb_clone_tx_timestamp(skb); 4384 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP) 4385 skb_tstamp_tx(skb, NULL); 4386 } 4387 4388 /** 4389 * skb_complete_wifi_ack - deliver skb with wifi status 4390 * 4391 * @skb: the original outgoing packet 4392 * @acked: ack status 4393 * 4394 */ 4395 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked); 4396 4397 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len); 4398 __sum16 __skb_checksum_complete(struct sk_buff *skb); 4399 4400 static inline int skb_csum_unnecessary(const struct sk_buff *skb) 4401 { 4402 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) || 4403 skb->csum_valid || 4404 (skb->ip_summed == CHECKSUM_PARTIAL && 4405 skb_checksum_start_offset(skb) >= 0)); 4406 } 4407 4408 /** 4409 * skb_checksum_complete - Calculate checksum of an entire packet 4410 * @skb: packet to process 4411 * 4412 * This function calculates the checksum over the entire packet plus 4413 * the value of skb->csum. The latter can be used to supply the 4414 * checksum of a pseudo header as used by TCP/UDP. It returns the 4415 * checksum. 4416 * 4417 * For protocols that contain complete checksums such as ICMP/TCP/UDP, 4418 * this function can be used to verify that checksum on received 4419 * packets. In that case the function should return zero if the 4420 * checksum is correct. In particular, this function will return zero 4421 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the 4422 * hardware has already verified the correctness of the checksum. 4423 */ 4424 static inline __sum16 skb_checksum_complete(struct sk_buff *skb) 4425 { 4426 return skb_csum_unnecessary(skb) ? 4427 0 : __skb_checksum_complete(skb); 4428 } 4429 4430 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb) 4431 { 4432 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4433 if (skb->csum_level == 0) 4434 skb->ip_summed = CHECKSUM_NONE; 4435 else 4436 skb->csum_level--; 4437 } 4438 } 4439 4440 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb) 4441 { 4442 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4443 if (skb->csum_level < SKB_MAX_CSUM_LEVEL) 4444 skb->csum_level++; 4445 } else if (skb->ip_summed == CHECKSUM_NONE) { 4446 skb->ip_summed = CHECKSUM_UNNECESSARY; 4447 skb->csum_level = 0; 4448 } 4449 } 4450 4451 static inline void __skb_reset_checksum_unnecessary(struct sk_buff *skb) 4452 { 4453 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4454 skb->ip_summed = CHECKSUM_NONE; 4455 skb->csum_level = 0; 4456 } 4457 } 4458 4459 /* Check if we need to perform checksum complete validation. 4460 * 4461 * Returns true if checksum complete is needed, false otherwise 4462 * (either checksum is unnecessary or zero checksum is allowed). 4463 */ 4464 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb, 4465 bool zero_okay, 4466 __sum16 check) 4467 { 4468 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) { 4469 skb->csum_valid = 1; 4470 __skb_decr_checksum_unnecessary(skb); 4471 return false; 4472 } 4473 4474 return true; 4475 } 4476 4477 /* For small packets <= CHECKSUM_BREAK perform checksum complete directly 4478 * in checksum_init. 4479 */ 4480 #define CHECKSUM_BREAK 76 4481 4482 /* Unset checksum-complete 4483 * 4484 * Unset checksum complete can be done when packet is being modified 4485 * (uncompressed for instance) and checksum-complete value is 4486 * invalidated. 4487 */ 4488 static inline void skb_checksum_complete_unset(struct sk_buff *skb) 4489 { 4490 if (skb->ip_summed == CHECKSUM_COMPLETE) 4491 skb->ip_summed = CHECKSUM_NONE; 4492 } 4493 4494 /* Validate (init) checksum based on checksum complete. 4495 * 4496 * Return values: 4497 * 0: checksum is validated or try to in skb_checksum_complete. In the latter 4498 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo 4499 * checksum is stored in skb->csum for use in __skb_checksum_complete 4500 * non-zero: value of invalid checksum 4501 * 4502 */ 4503 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb, 4504 bool complete, 4505 __wsum psum) 4506 { 4507 if (skb->ip_summed == CHECKSUM_COMPLETE) { 4508 if (!csum_fold(csum_add(psum, skb->csum))) { 4509 skb->csum_valid = 1; 4510 return 0; 4511 } 4512 } 4513 4514 skb->csum = psum; 4515 4516 if (complete || skb->len <= CHECKSUM_BREAK) { 4517 __sum16 csum; 4518 4519 csum = __skb_checksum_complete(skb); 4520 skb->csum_valid = !csum; 4521 return csum; 4522 } 4523 4524 return 0; 4525 } 4526 4527 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto) 4528 { 4529 return 0; 4530 } 4531 4532 /* Perform checksum validate (init). Note that this is a macro since we only 4533 * want to calculate the pseudo header which is an input function if necessary. 4534 * First we try to validate without any computation (checksum unnecessary) and 4535 * then calculate based on checksum complete calling the function to compute 4536 * pseudo header. 4537 * 4538 * Return values: 4539 * 0: checksum is validated or try to in skb_checksum_complete 4540 * non-zero: value of invalid checksum 4541 */ 4542 #define __skb_checksum_validate(skb, proto, complete, \ 4543 zero_okay, check, compute_pseudo) \ 4544 ({ \ 4545 __sum16 __ret = 0; \ 4546 skb->csum_valid = 0; \ 4547 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \ 4548 __ret = __skb_checksum_validate_complete(skb, \ 4549 complete, compute_pseudo(skb, proto)); \ 4550 __ret; \ 4551 }) 4552 4553 #define skb_checksum_init(skb, proto, compute_pseudo) \ 4554 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo) 4555 4556 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \ 4557 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo) 4558 4559 #define skb_checksum_validate(skb, proto, compute_pseudo) \ 4560 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo) 4561 4562 #define skb_checksum_validate_zero_check(skb, proto, check, \ 4563 compute_pseudo) \ 4564 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo) 4565 4566 #define skb_checksum_simple_validate(skb) \ 4567 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo) 4568 4569 static inline bool __skb_checksum_convert_check(struct sk_buff *skb) 4570 { 4571 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid); 4572 } 4573 4574 static inline void __skb_checksum_convert(struct sk_buff *skb, __wsum pseudo) 4575 { 4576 skb->csum = ~pseudo; 4577 skb->ip_summed = CHECKSUM_COMPLETE; 4578 } 4579 4580 #define skb_checksum_try_convert(skb, proto, compute_pseudo) \ 4581 do { \ 4582 if (__skb_checksum_convert_check(skb)) \ 4583 __skb_checksum_convert(skb, compute_pseudo(skb, proto)); \ 4584 } while (0) 4585 4586 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr, 4587 u16 start, u16 offset) 4588 { 4589 skb->ip_summed = CHECKSUM_PARTIAL; 4590 skb->csum_start = ((unsigned char *)ptr + start) - skb->head; 4591 skb->csum_offset = offset - start; 4592 } 4593 4594 /* Update skbuf and packet to reflect the remote checksum offload operation. 4595 * When called, ptr indicates the starting point for skb->csum when 4596 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete 4597 * here, skb_postpull_rcsum is done so skb->csum start is ptr. 4598 */ 4599 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr, 4600 int start, int offset, bool nopartial) 4601 { 4602 __wsum delta; 4603 4604 if (!nopartial) { 4605 skb_remcsum_adjust_partial(skb, ptr, start, offset); 4606 return; 4607 } 4608 4609 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) { 4610 __skb_checksum_complete(skb); 4611 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data); 4612 } 4613 4614 delta = remcsum_adjust(ptr, skb->csum, start, offset); 4615 4616 /* Adjust skb->csum since we changed the packet */ 4617 skb->csum = csum_add(skb->csum, delta); 4618 } 4619 4620 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb) 4621 { 4622 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 4623 return (void *)(skb->_nfct & NFCT_PTRMASK); 4624 #else 4625 return NULL; 4626 #endif 4627 } 4628 4629 static inline unsigned long skb_get_nfct(const struct sk_buff *skb) 4630 { 4631 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 4632 return skb->_nfct; 4633 #else 4634 return 0UL; 4635 #endif 4636 } 4637 4638 static inline void skb_set_nfct(struct sk_buff *skb, unsigned long nfct) 4639 { 4640 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 4641 skb->slow_gro |= !!nfct; 4642 skb->_nfct = nfct; 4643 #endif 4644 } 4645 4646 #ifdef CONFIG_SKB_EXTENSIONS 4647 enum skb_ext_id { 4648 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 4649 SKB_EXT_BRIDGE_NF, 4650 #endif 4651 #ifdef CONFIG_XFRM 4652 SKB_EXT_SEC_PATH, 4653 #endif 4654 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 4655 TC_SKB_EXT, 4656 #endif 4657 #if IS_ENABLED(CONFIG_MPTCP) 4658 SKB_EXT_MPTCP, 4659 #endif 4660 #if IS_ENABLED(CONFIG_MCTP_FLOWS) 4661 SKB_EXT_MCTP, 4662 #endif 4663 SKB_EXT_NUM, /* must be last */ 4664 }; 4665 4666 /** 4667 * struct skb_ext - sk_buff extensions 4668 * @refcnt: 1 on allocation, deallocated on 0 4669 * @offset: offset to add to @data to obtain extension address 4670 * @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units 4671 * @data: start of extension data, variable sized 4672 * 4673 * Note: offsets/lengths are stored in chunks of 8 bytes, this allows 4674 * to use 'u8' types while allowing up to 2kb worth of extension data. 4675 */ 4676 struct skb_ext { 4677 refcount_t refcnt; 4678 u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */ 4679 u8 chunks; /* same */ 4680 char data[] __aligned(8); 4681 }; 4682 4683 struct skb_ext *__skb_ext_alloc(gfp_t flags); 4684 void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id, 4685 struct skb_ext *ext); 4686 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id); 4687 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id); 4688 void __skb_ext_put(struct skb_ext *ext); 4689 4690 static inline void skb_ext_put(struct sk_buff *skb) 4691 { 4692 if (skb->active_extensions) 4693 __skb_ext_put(skb->extensions); 4694 } 4695 4696 static inline void __skb_ext_copy(struct sk_buff *dst, 4697 const struct sk_buff *src) 4698 { 4699 dst->active_extensions = src->active_extensions; 4700 4701 if (src->active_extensions) { 4702 struct skb_ext *ext = src->extensions; 4703 4704 refcount_inc(&ext->refcnt); 4705 dst->extensions = ext; 4706 } 4707 } 4708 4709 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src) 4710 { 4711 skb_ext_put(dst); 4712 __skb_ext_copy(dst, src); 4713 } 4714 4715 static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i) 4716 { 4717 return !!ext->offset[i]; 4718 } 4719 4720 static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id) 4721 { 4722 return skb->active_extensions & (1 << id); 4723 } 4724 4725 static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id) 4726 { 4727 if (skb_ext_exist(skb, id)) 4728 __skb_ext_del(skb, id); 4729 } 4730 4731 static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id) 4732 { 4733 if (skb_ext_exist(skb, id)) { 4734 struct skb_ext *ext = skb->extensions; 4735 4736 return (void *)ext + (ext->offset[id] << 3); 4737 } 4738 4739 return NULL; 4740 } 4741 4742 static inline void skb_ext_reset(struct sk_buff *skb) 4743 { 4744 if (unlikely(skb->active_extensions)) { 4745 __skb_ext_put(skb->extensions); 4746 skb->active_extensions = 0; 4747 } 4748 } 4749 4750 static inline bool skb_has_extensions(struct sk_buff *skb) 4751 { 4752 return unlikely(skb->active_extensions); 4753 } 4754 #else 4755 static inline void skb_ext_put(struct sk_buff *skb) {} 4756 static inline void skb_ext_reset(struct sk_buff *skb) {} 4757 static inline void skb_ext_del(struct sk_buff *skb, int unused) {} 4758 static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {} 4759 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {} 4760 static inline bool skb_has_extensions(struct sk_buff *skb) { return false; } 4761 #endif /* CONFIG_SKB_EXTENSIONS */ 4762 4763 static inline void nf_reset_ct(struct sk_buff *skb) 4764 { 4765 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 4766 nf_conntrack_put(skb_nfct(skb)); 4767 skb->_nfct = 0; 4768 #endif 4769 } 4770 4771 static inline void nf_reset_trace(struct sk_buff *skb) 4772 { 4773 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES) 4774 skb->nf_trace = 0; 4775 #endif 4776 } 4777 4778 static inline void ipvs_reset(struct sk_buff *skb) 4779 { 4780 #if IS_ENABLED(CONFIG_IP_VS) 4781 skb->ipvs_property = 0; 4782 #endif 4783 } 4784 4785 /* Note: This doesn't put any conntrack info in dst. */ 4786 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src, 4787 bool copy) 4788 { 4789 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 4790 dst->_nfct = src->_nfct; 4791 nf_conntrack_get(skb_nfct(src)); 4792 #endif 4793 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES) 4794 if (copy) 4795 dst->nf_trace = src->nf_trace; 4796 #endif 4797 } 4798 4799 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src) 4800 { 4801 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 4802 nf_conntrack_put(skb_nfct(dst)); 4803 #endif 4804 dst->slow_gro = src->slow_gro; 4805 __nf_copy(dst, src, true); 4806 } 4807 4808 #ifdef CONFIG_NETWORK_SECMARK 4809 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 4810 { 4811 to->secmark = from->secmark; 4812 } 4813 4814 static inline void skb_init_secmark(struct sk_buff *skb) 4815 { 4816 skb->secmark = 0; 4817 } 4818 #else 4819 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 4820 { } 4821 4822 static inline void skb_init_secmark(struct sk_buff *skb) 4823 { } 4824 #endif 4825 4826 static inline int secpath_exists(const struct sk_buff *skb) 4827 { 4828 #ifdef CONFIG_XFRM 4829 return skb_ext_exist(skb, SKB_EXT_SEC_PATH); 4830 #else 4831 return 0; 4832 #endif 4833 } 4834 4835 static inline bool skb_irq_freeable(const struct sk_buff *skb) 4836 { 4837 return !skb->destructor && 4838 !secpath_exists(skb) && 4839 !skb_nfct(skb) && 4840 !skb->_skb_refdst && 4841 !skb_has_frag_list(skb); 4842 } 4843 4844 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping) 4845 { 4846 skb->queue_mapping = queue_mapping; 4847 } 4848 4849 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb) 4850 { 4851 return skb->queue_mapping; 4852 } 4853 4854 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from) 4855 { 4856 to->queue_mapping = from->queue_mapping; 4857 } 4858 4859 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue) 4860 { 4861 skb->queue_mapping = rx_queue + 1; 4862 } 4863 4864 static inline u16 skb_get_rx_queue(const struct sk_buff *skb) 4865 { 4866 return skb->queue_mapping - 1; 4867 } 4868 4869 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb) 4870 { 4871 return skb->queue_mapping != 0; 4872 } 4873 4874 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val) 4875 { 4876 skb->dst_pending_confirm = val; 4877 } 4878 4879 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb) 4880 { 4881 return skb->dst_pending_confirm != 0; 4882 } 4883 4884 static inline struct sec_path *skb_sec_path(const struct sk_buff *skb) 4885 { 4886 #ifdef CONFIG_XFRM 4887 return skb_ext_find(skb, SKB_EXT_SEC_PATH); 4888 #else 4889 return NULL; 4890 #endif 4891 } 4892 4893 static inline bool skb_is_gso(const struct sk_buff *skb) 4894 { 4895 return skb_shinfo(skb)->gso_size; 4896 } 4897 4898 /* Note: Should be called only if skb_is_gso(skb) is true */ 4899 static inline bool skb_is_gso_v6(const struct sk_buff *skb) 4900 { 4901 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6; 4902 } 4903 4904 /* Note: Should be called only if skb_is_gso(skb) is true */ 4905 static inline bool skb_is_gso_sctp(const struct sk_buff *skb) 4906 { 4907 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP; 4908 } 4909 4910 /* Note: Should be called only if skb_is_gso(skb) is true */ 4911 static inline bool skb_is_gso_tcp(const struct sk_buff *skb) 4912 { 4913 return skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6); 4914 } 4915 4916 static inline void skb_gso_reset(struct sk_buff *skb) 4917 { 4918 skb_shinfo(skb)->gso_size = 0; 4919 skb_shinfo(skb)->gso_segs = 0; 4920 skb_shinfo(skb)->gso_type = 0; 4921 } 4922 4923 static inline void skb_increase_gso_size(struct skb_shared_info *shinfo, 4924 u16 increment) 4925 { 4926 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS)) 4927 return; 4928 shinfo->gso_size += increment; 4929 } 4930 4931 static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo, 4932 u16 decrement) 4933 { 4934 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS)) 4935 return; 4936 shinfo->gso_size -= decrement; 4937 } 4938 4939 void __skb_warn_lro_forwarding(const struct sk_buff *skb); 4940 4941 static inline bool skb_warn_if_lro(const struct sk_buff *skb) 4942 { 4943 /* LRO sets gso_size but not gso_type, whereas if GSO is really 4944 * wanted then gso_type will be set. */ 4945 const struct skb_shared_info *shinfo = skb_shinfo(skb); 4946 4947 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 && 4948 unlikely(shinfo->gso_type == 0)) { 4949 __skb_warn_lro_forwarding(skb); 4950 return true; 4951 } 4952 return false; 4953 } 4954 4955 static inline void skb_forward_csum(struct sk_buff *skb) 4956 { 4957 /* Unfortunately we don't support this one. Any brave souls? */ 4958 if (skb->ip_summed == CHECKSUM_COMPLETE) 4959 skb->ip_summed = CHECKSUM_NONE; 4960 } 4961 4962 /** 4963 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE 4964 * @skb: skb to check 4965 * 4966 * fresh skbs have their ip_summed set to CHECKSUM_NONE. 4967 * Instead of forcing ip_summed to CHECKSUM_NONE, we can 4968 * use this helper, to document places where we make this assertion. 4969 */ 4970 static inline void skb_checksum_none_assert(const struct sk_buff *skb) 4971 { 4972 DEBUG_NET_WARN_ON_ONCE(skb->ip_summed != CHECKSUM_NONE); 4973 } 4974 4975 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off); 4976 4977 int skb_checksum_setup(struct sk_buff *skb, bool recalculate); 4978 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 4979 unsigned int transport_len, 4980 __sum16(*skb_chkf)(struct sk_buff *skb)); 4981 4982 /** 4983 * skb_head_is_locked - Determine if the skb->head is locked down 4984 * @skb: skb to check 4985 * 4986 * The head on skbs build around a head frag can be removed if they are 4987 * not cloned. This function returns true if the skb head is locked down 4988 * due to either being allocated via kmalloc, or by being a clone with 4989 * multiple references to the head. 4990 */ 4991 static inline bool skb_head_is_locked(const struct sk_buff *skb) 4992 { 4993 return !skb->head_frag || skb_cloned(skb); 4994 } 4995 4996 /* Local Checksum Offload. 4997 * Compute outer checksum based on the assumption that the 4998 * inner checksum will be offloaded later. 4999 * See Documentation/networking/checksum-offloads.rst for 5000 * explanation of how this works. 5001 * Fill in outer checksum adjustment (e.g. with sum of outer 5002 * pseudo-header) before calling. 5003 * Also ensure that inner checksum is in linear data area. 5004 */ 5005 static inline __wsum lco_csum(struct sk_buff *skb) 5006 { 5007 unsigned char *csum_start = skb_checksum_start(skb); 5008 unsigned char *l4_hdr = skb_transport_header(skb); 5009 __wsum partial; 5010 5011 /* Start with complement of inner checksum adjustment */ 5012 partial = ~csum_unfold(*(__force __sum16 *)(csum_start + 5013 skb->csum_offset)); 5014 5015 /* Add in checksum of our headers (incl. outer checksum 5016 * adjustment filled in by caller) and return result. 5017 */ 5018 return csum_partial(l4_hdr, csum_start - l4_hdr, partial); 5019 } 5020 5021 static inline bool skb_is_redirected(const struct sk_buff *skb) 5022 { 5023 return skb->redirected; 5024 } 5025 5026 static inline void skb_set_redirected(struct sk_buff *skb, bool from_ingress) 5027 { 5028 skb->redirected = 1; 5029 #ifdef CONFIG_NET_REDIRECT 5030 skb->from_ingress = from_ingress; 5031 if (skb->from_ingress) 5032 skb_clear_tstamp(skb); 5033 #endif 5034 } 5035 5036 static inline void skb_reset_redirect(struct sk_buff *skb) 5037 { 5038 skb->redirected = 0; 5039 } 5040 5041 static inline void skb_set_redirected_noclear(struct sk_buff *skb, 5042 bool from_ingress) 5043 { 5044 skb->redirected = 1; 5045 #ifdef CONFIG_NET_REDIRECT 5046 skb->from_ingress = from_ingress; 5047 #endif 5048 } 5049 5050 static inline bool skb_csum_is_sctp(struct sk_buff *skb) 5051 { 5052 #if IS_ENABLED(CONFIG_IP_SCTP) 5053 return skb->csum_not_inet; 5054 #else 5055 return 0; 5056 #endif 5057 } 5058 5059 static inline void skb_reset_csum_not_inet(struct sk_buff *skb) 5060 { 5061 skb->ip_summed = CHECKSUM_NONE; 5062 #if IS_ENABLED(CONFIG_IP_SCTP) 5063 skb->csum_not_inet = 0; 5064 #endif 5065 } 5066 5067 static inline void skb_set_kcov_handle(struct sk_buff *skb, 5068 const u64 kcov_handle) 5069 { 5070 #ifdef CONFIG_KCOV 5071 skb->kcov_handle = kcov_handle; 5072 #endif 5073 } 5074 5075 static inline u64 skb_get_kcov_handle(struct sk_buff *skb) 5076 { 5077 #ifdef CONFIG_KCOV 5078 return skb->kcov_handle; 5079 #else 5080 return 0; 5081 #endif 5082 } 5083 5084 static inline void skb_mark_for_recycle(struct sk_buff *skb) 5085 { 5086 #ifdef CONFIG_PAGE_POOL 5087 skb->pp_recycle = 1; 5088 #endif 5089 } 5090 5091 ssize_t skb_splice_from_iter(struct sk_buff *skb, struct iov_iter *iter, 5092 ssize_t maxsize, gfp_t gfp); 5093 5094 #endif /* __KERNEL__ */ 5095 #endif /* _LINUX_SKBUFF_H */ 5096